> Here is the state of venture capital in early 2025: Venture capital is moribund except AI. AI is moribund except OpenAI. OpenAI is a weird scam that wants to burn money so fast it summons AI God. Nobody can cash out.
The interesting thing, to me, is how speculative OpenAI's bet is.
IIRC it was 2019 when I tinkered with the versions of GPT 2.0 that had web interfaces, and they were interesting toys. Then I began using ChatGPT since its launch, which was around Dec 2022, and that was a profound paradigm shift. It showed real emergent behavior and it was capable of very interesting things.
2019 - 2022 was three years. No hype, no trillions of dollars invested, but tremendous progress.
Now, there has been progress in the part ~three years in synthetic benchmarks, but the feeling with ChatGPT 4.5 today is still the same as it was with GPT-3/GPT-4 in 2022. 4.5/o3 doesn't seem hugely more intelligent then 3.0 -- it hallucinates less, and it's capable of running web searches and doing directed research -- but it's no paradigm shift. If things keep progressing the way they're going, we'll get better interfaces and more tools, but it's far from clear that superintelligence (more-than-human insight, skill, and inventiveness,) is even possible with LLMs.
- Have a real time video conversation with an AI which can see what you see, translate between languages, read text, recognize objects, and interact with the real world.
Contrary to your (potentially misremembered?) history, nothing at all like this was possible in 2019. I remember finetuning an early GPT-2 (before they even released the 2B model!) on a large corpus of Star Wars novels and being impressed that it would mention "Luke" when I ran the produced model! Now I wear it on my head and read restaurant menus with it. Use it to find my Uber (what kind of car is that?) Today I am building my raised garden beds out back and reading the various soil amendments I purchased, talking about how much bloodmeal to put over the hugelkultur layer, having it do math, and generally having a pair of eyeballs. I'm blind. The amount of utility I get out of these things is ... very hard to overstate.
> - Have a real time video conversation with an AI which can see what you see, translate between languages, read text, recognize objects, and interact with the real world.
Maybe it's me having an extremely low imagination, but that stuff existed for a while in the shape of google lens and the various vision flavor of LLMs, and I must have used them.... 3 times in years, and not once did I think "Gosh I wish I could just ask a question aloud while walking in the street about this building and wait for the answer". It's either important enough that I want to see the wikipedia page straight from google maps and read the whole lot or not.
> an AI which can read text, recognize objects, and interact with the real world.
I can already do that pretty well with my eyeballs, and I don't need to worry about hallucinations, privacy, bad phone signal or my bad english accent. I get that is certainly an amazing tools for people with vision impairments, but that is not the market Meta/OpenAI are aiming for and forcefully trying to shove it into.
So yes, mayyybe if I am in a foreign country I could see a use but I usually want to get _away_ from technology on vacation. So I really don't see the point, but it seems that they believe I am the target audience?
Am I mistaken in thinking that much of what you're describing would be considered computer vision, and that computer vision was already largely capable of these things in 2019 and before? I vividly remember a live demonstration of an on-device AR-and-object-recognition program at the 2014 Facebook developer conference.
> Ask to play with a pear [sic] of the Meta glasses.
Ironically, this typo is very likely a result of AI dictation making a mistake. There are a lot of common misspellings in English, like "their" and "there", but I've never seen a human confuse "pair" and "pear".
So yeah, there are cool demos you can do that you couldn't five years ago. But whether any of those cool demos actually translate into something useful in day-to-day life where the benefits outweigh the costs and risks is far from clear.
One thing that’s fascinating to me is that these straight out of sci-fi things are novelties or demos, and don’t seem all that popular. Most people just aren’t interested.
It’s the “boring” stuff that’s interesting: automating drudgery work, a better way to research, etc.
I’ve been predicting for years that glasses — whether AR or VR — are and will remain niche. I don’t think most people want them.
For a long time I held a sentiment similar to the parent comment, and then my brother sat me down, took out his phone, put chatGPT into conversation mode, and chatted with it for about five minutes. That was the second time I was truly amazed by chatGPT (after my first conversation, where I got it to tell me a fair bit about how Postgres works). Its ability to hold a natural, context-aware conversation has gotten really amazing.
I somehow agree with the op, that I don’t think I’m much closer to hiring chatGPT for a real job in 2025 than I was in 2022, but also you that there has been meaningful progress. And in particular, products that are transformative for disabled people are usually big improvements to the status quo for abled people too (oxo good grips being the classic example—transformative for people with arthritis, and generally just better for everybody else)
And we had these toys with higher latency in 22 with gpt3. The better tooling and integration hides the extremely slowed down pace of innovation in base models all while throwing mountains of compute at it.
The comment quotes the article as stating that (so-called "tech") "[v]enture capital is moribund except for AI. AI is moribund except OpenAI."
How does this personal anecdote relate to the observation that (so-called "tech") "[v]enture capital is moribund except for AI. AI is moribund except OpenAI."
Is venture capital involved with the Meta Ray-Ban glasses that we are advised to try for free.
> - Have a real time video conversation with an AI which can see what you see, translate between languages, read text, recognize objects, and interact with the real world.
This certainly provides benefit to those with limited vision, which is great. But that is a very small segment of consumers. Besides those, how many other people do you know who are actually _using these glasses_ in the real world?
The Facebook glasses aren't as relevant as the LLMs doing many thinking tasks that people get paid to do. A lot of it is as a productivity multiplier, and I'm not saying it's all doom and gloom, but it's transformative.
This is sort of why I say that the hype is ultimately detrimental to the healthy development of tech.
Generative AI was a sort of paradigm shift, and can be developed into interesting tools that boost human productivity. But those things take time, sometimes decades to reach maturity.
That is not good for the get rich quick machine of Venture Capital and Hustle Culture, where quick exits require a bunch of bag holders.
You gotta have suckers, and for that Gen AI cannot be an "interesting technology" with good potential. It needs to be "the future", that will disrupt everything and everyone.
> 4.5/o3 doesn't seem hugely more intelligent then 3.0 -- it hallucinates less [...]
This is not entirely true, or at least the trend is not necessarily less hallucination. See section 3.3 in the OpenAI o3 and o4-mini System Card[1], which shows that o3 and o4-mini both hallucinate more than o1. See also [2] for more data on hallucinations.
does it? An anecdote from yesterday: My wife was asked to do a lit review as an assignment for nursing school. Her professor sent her an example of papers on the topic, with a brief "relevance" summary for each. My wife asked me for help as she was frustrated she couldn't find any of the referenced papers online (she's not the most adept at technology and figured she was doing something wrong). I took one look at the email from her professor and could tell just by the formatting that it was LLM generated (which model, I don't know, but obviously a 2025 model). The professor didn't say anything about using an LLM, and my wife didn't suspect that might be the case.
My wife and I did some Google Scholar searches, and _every_ _single_ _one_ of the 5 papers cited did _not_ exist. In 2 of the cases, similar papers did exist, but with different authors or a different title that resembled the fake "citation". The three others did not exist in any form - there were other papers on the same subject, sure, but nothing closely resembling the "citations" either in terms of authorship or title.
The global broadband network that is the physical internet took about 10 years and $1T to build, mostly between 1998 and 2008 and it made the internet massively better every year of the build-out. That's also precisely as much time and money as has been put into this generative AI bubble.
The internet was adding a trillion dollars to the global economy by 2008 and the end of that rapid expansion, where AI is still sucking hundreds of billions a year into a black hole with no killer use cases that could possibly pay off its investment, much less begin adding trillions to the global economy.
And a decade before the web and internet explosion, PCs were similar, with a massive build out and immediate massive returns.
This excuse making for AI is getting old. It needs to put up or shut up because so far it's a joke compared to real advances like the PC and the Internet, all while being hyped by VC collecting companies as the arrival of a literal God.
We look at CPUs or the transmission of digital data and these seem to have improved exponentially but these are rather exceptions and are composed of multiple technologies at different stages. Like how we went from internet through phone lines, to dedicated copper lines for data, to optic fiber straight into to people's homes.
Eg: look how the efficiency of solar cells has progressed over the last 50 years
True, but the religion of the Singularity that fueled this round of financing was premised on this improvement growing exponentially fast, thanks to the support provided by the current version of AI tools. There's no sign this will happen anytime soon.
I think you mean "invention". Innovation describes how products change over time and doesn't necessarily imply insight or value-adds. Sometimes it just means the packaging gets updated.
Agreed. It's become a pretty useful tool for individual people to use for individual tasks. But to hyper-scale to a point where it justifies crazy valuations - even short of super-intelligence - it has to scale beyond users. It has to no longer require human oversight from one step to the next.
I think that's the key threshold all these companies have been running up against, and crossing it would be the paradigm shift we keep hearing about. But they've been trying for years, and haven't done it yet, and seem to be plateauing
And then in OpenAI's case specifically- this tech has become commoditized really quickly. They have several direct competitors, including a few that are open, and their only real moat is their brand and Sam's fundraising ability. Their UX is one of the best right now, but that isn't really a moat
What? MSFT is worth 3.25 trillion without scaling beyond any users. All these AI companies need is for everyone to pay them $100 a year for a subscription and they have more than justified their valuation. Same strategy Microsoft uses
> ChatGPT 4.5 today is still the same as it was with GPT-3/GPT-4 in 2022. 4.5/o3 doesn't seem hugely more intelligent then 3.0
I think you're misremembering how 3.0 worked. Granted, the slope from 2.0 to 3.0 was very steep, but a ton of progress has happened in the past few years.
Agreed, and as someone who's used it for work most days since 3.0 launch, it's likely way more efficient in outcomes - maybe 30-40%. But as the GP said, there's no paradigm shift. All the new features feel very gimicky, and OpenAI lost their first mover advantage quite a while ago.
Yea it hallucinates less, but it still hallucinates a lot. I think we’re proving that intelligence is not just a language model, even an obscenely large one.
It's actually hallucinating more. The more they mess with these now mature models, models which no longer scale with training corpus size, the more they push and prod at the models to do better with chain of thought and other techniques, the more they hallucinate. It's getting worse because the magic of LLM scaling is over and the techniques we have to make them better actually make the hallucinations worse.
I'm waiting impatiently for someone to connect the obvious dots and roll out an AI assistant intentionally patterned after Hunter S Thompson. The hallucinations are now a feature.
I mean, if you look up the data around eye witness testimony being extremely unreliable even for people who directly witnessed things, it's not out of the realm that humans regularly hallucinate as well
I must say I do feel the newest GPT Versions are vastly better than the old ones. I found GPT-3 Stuff to be an interesting toy but too often it was too wrong and too stubborn to be useful. I use the 4.0+ Version regularly know.
Just recently I took a screenshot from a jira Burndown chart to write a description of the sprint progress for our stakeholders. Did it in one shot from a screenshot and got it right.
I think it's simplification to compare progress only on LLM level.
We had big progress in AI in last 2 years but have to take into account more than text token generation. We have image generation that is not only super realistic but you just text what you want to modify without learning complicated tools like ComfyUI.
We have text to speech and audio to audio that is not only very realistic and fluent with many languages but also can express emotions in speech.
We have video generation that is really more realistic every month and taking less computation.
There is big progress in 3d models generation. Speech to text is still getting improved and fast enough to run on phones reducing latency. Next frontier is how AI is applied for robotics. No to mention areas not sexy to end users but in application in healthcare.
I have a similar feeling. While LLMs have given me a new way to do search/questions, it is the byproducts that feel like the actual game changers. For me, it is vision models and pretty impressive STT and TTS. I am blind, so I have my own reasons why Vision and Speech have so many real world applications for me. Sure, LLMs are still the backbone of the applications emerging, but the real progress in terms of use cases is in the fringes.
Heck, I wrote myself my own personal radio moderator in a few hundred lines of shell, later rewritten in Python. As a simple MPD client. Watch out for a queued track which has albumart, and pass the track metadata + picture to the LLM. Send the result through a pretty natural sounding TTS, and queue the resulting sound file before the next track. Suddenly, I had a radio moderator that would narrate album art for me. It gave me a glimpse into a world that wouldn't have been possible before. And while the LLM is basically writing the script, the real magic comes from multimodal and great sounding TTS.
Much potential for really cool looking/sounding PoCs. However, what makes me worry is that there is not much progress on (to me) obvious shortcomings. For instance, OpenAI TTS really can't speak any numbers correctly. Digits maybe, but once you hand it something like "2025" the chance is high it will have pronounciation problems. In the first months, this felt like bad but temporarily acceptable. A year later, it feels like hilariously sad that nothing has been done to address such a simple yet important issue. You know that something bad is going on when you start to consider expanding numbers to written-out form before passing the message to the TTS. My girlfriend keeps joking that since LLMs, we now have computers that totally can not compute correctly. And she has a point. Sure, hand the LLM a tool to do calculations, and the situation improves somewhat. But it seems to be underlying, as shown by the problems of TTS.
Vision models have so many applications for me... However, some of them turn out to be actually unusable in practice. That becomes clear when you use a vision model to read the values off a blood pressure sensor. Take three photos, and you get three slightly different values. Not obviously made up stuff, but numbers that could be. 145/90, 147/93, 142/97. Well, the range might be clear, but actually, you can never be sure. Great for scene and art descriptions, since hallucinations almost fall through the cracks. But I would never use it to read any kind of data, neither OCR'd text nor, gasp, numbers! You can never know if you have been lied to.
But still, some of the byproducts of LLMs feel like a real revolution. The moment you realize why whisper is named like that. When you test it on your laptop, and realize that it just transcribed the YouTube video you were rather silently running in the background. Some of this stuff feels like a big jump.
> 4.5/o3 doesn't seem hugely more intelligent then 3.0
I disagree with 3.0, but perhaps that feels true for 4.0 or even 3.5 for some queries.
The reason is that when LLMs are asked questions whose answers can be interpolated or retrieved from their training data, they will likely use widely accepted human knowledge or patterns to compose their responses. (This is a simplification of how LLMs work, just to illustrate the key point here.) This knowledge has been refined and has evolved through decades of human experiments and experiences.
Domain experts of varying intelligence will likely come up with similar replies on these largely routine questions as well.
The difference shows up when you pose a query that demands deep reasoning or requires expertise in multiple fields. Then, frontier reasoning models like o3 can sometimes form creative solutions that are not textbook answers.
I strongly suspect that Reinforcement Learning with feedback from high-quality simulations or real environments will be key for these models' capabilities to surpass those of human experts.
Superhuman milestones, equivalent to those achieved by AlphaGo and AlphaZero between 2016 and 2018, might be reached in several fields over the coming years. This will likely happen first in fields with rapid feedback loops and highly accurate simulators, e.g. math problem solving (as opposed to novel mathematical research), coding (as opposed to product innovation).
LLMs are in the limelight, but I won't dismiss the notable progress in areas like vision and visual parsing, and also image synthesis and transformation. Robotic taxis finally are running down city streets, and drive on par with humans, or better. You can give the machine a rough sketch, and get a well-made picture. You can show the machine a few photos, and get a very reasonable 3D model, even in a form of reasonable meshes. Etc.
> The interesting thing, to me, is how speculative OpenAI's bet is.
It doesn't really matter how speculative the AGI bet is, their consumer AI business by itself is basically guaranteed to drown them in money. The only reason they're making losses at the moment is because they're choosing not to monetize their free tier users with ads, presumably since they don't need to make a profit and can prioritize growth.
But the moment they flip the advertising switch, their traffic will be both highly monetizable and ludicrously high margin.
1. People won’t ultimate go download ChatGPT.app or use a website — they’re going to be using the functionality through structured services in iOS and Android, and it’s necessarily going to be under control of the OS vendors for security/privacy reasons. This doesn’t mean Apple and Google own the LLMs — there will be consumer choice for antitrust reasons if nothing else — but the operating system has to be a conduit for access to your data, and also for unified user experience. Which means advertising will be limited.
2. Say it does go the way you think it will — what prevents a real non-profit, open source LLM from taking it away from the commercial players? There really is no moat (other than money, energy, data center space).
Not really. The best math I've seen says OAI cannot break even with ads on free tiers without about 5X as many users. That's to break even, not to be wildly profitable as you seem to be suggesting. OAI will need more than ChatGPT, pro and "free" to be anything close to web search for ad revenue.
+ one thing I just absolutely can‘t understand: What‘s their moat? Their offerings are replaceable by like 5 direct competitors, some of which are open source. Pricewise, it‘s a race to the bottom, and the bottom is „api pricing vs running open source yourself“.
I think it’s worth mentioning that OpenAI was founded in 2015, so ChatGPT's overnight success in 2022 was 7 years in the making. Looking at that timescale,
I'd say Google counts, as do several other companies is we look at inception till successful product.
I totally disagree. The initial release of ChatGPT felt magical because of how cool was to be able to hold a natural, fluid conversation with a computer. But it was an extremely stupid computer beneath that veneer of conversational intelligence. Recent models can do a lot of math accurately without “reasoning” or invoking tools; GPT-3 could barely perform basic arithmetic.
And the latest models get beat on reliability by a $2.99 drug store slim wallet sized solar powered calculator you could have bought 35 years ago for about that same price.
I agree with your take, but at the same time, I’m not sure these things should ever do math. I know that they can, but it seems impossible to draw the line of math they can do vs math they shouldn’t do. A part of me suspects they should always be outsourcing any math to a tool.
I understand the spirit in this line of criticism, but I think it's easy to muddle the timelines and feel as if things "aren't moving," when in fact, the pace of research and improvement is great.
For context:
- GPT 2 was released in Feb 2019
- GPT 3 came out roughly 18 months later in 2020. It was a huge jump, but still not "usable" for many things.
- InstructGPT came out roughly 18 months later in early 2022, and was a huge advancement. This is RLHF's big moment.
- About 10 months later, ChatGPT is released at the end of 2022 as a "sibling" to InstructGPT. It's an "open research preview" at this point. This is around the time OpenAI starts referring to certain models as being in the "3.5 family"
- GPT-4 comes out in March 2023, so barely 2 years ago now. Huge jumps in performance, context window size, and it supports images. This is around the time ChatGPT hits 100 million users and is really becoming a reliable, widely adopted tool. This is also the same time that tools like Cursor are hitting the market, though they haven't exploded yet. Models are just now getting "good enough" for these kinds of applications
- GPT-4-Turbo comes out in November 2023, with way larger context windows and lower pricing.
- About 12 months ago, GPT-4o released, showing slightly increased performance on existing benchmarks over 4, but now with state-of-the-art audio capability support for something like 50 languages.
- 5 months ago, o1 releases. This is a big moment for scaling compute at test time, which is a major current research direction in ML. Shows huge improvements (something like 8x over 4) on some math/reasoning benchmarks. Within months, we have o3 and o4, which substantially improve these scores even further.
- In February of this year, we get 4.5, and then months later, the confusingly named 4.1, which shows improvements over 4o.
So to be clear, in 2019 we had an interesting research project that only a few people could tinker with.
18 months later, we had a better model that you could play with via an API, but was still a toy.
It takes more than two years to go from that to ChatGPT, and a few more months (nearly 3 years total) to get to the "useful" version of ChatGPT that really sets the world on fire. It took roughly 4 + 1/2 years to go from "novelty text generation" to "useful text generation".
In the 2 years since then, we've gotten multimodal models, a new class of reasoning models, baseline improvement across performance, and more. If anything, there is more fundamental research and wider variety of directions now (the kind of stuff that shifts paradigms) than before.
I don't agree. I think it's still just a toy that can't get anything useful done. It still hallucinates quite a bit, and gives confidently wrong answers all the time. It's impressive that we could make a program that does all those things, but until they get the accuracy to a reasonable level it won't be useful.
I've got AI fatigue as the next guy, but this is over correcting.
VC money is in a constant state of FOMO, this is nothing new. Companies dress up as AI, or web3 or web2, or fintech or what ever to more easily attract capital. If 57.9% of dollars went to AI startups this year is not because everything is AI, I would bet 25% are just companies that tacked AI on to an unrelated business model and its skewing the statistics. I'll promise you that 10 years from now 57.9% of VC funding is going to be in some other buzzword and its not going to be AI.
Can confirm. I’m at a startup in a pretty boring space and we’re having lots of success just by bringing modern software into the industry, but we’re gearing up for a series A and we absolutely must include AI in our pitch deck. So for the last 6-8 months we’ve just been scrambling to find some sort of reasonable use cases for AI in our product, but in reality it’s not a differentiator for us at all.
If you can find nothing else, you can add a little irritating floating chatbot icon on your product(I am making huge assumptions that it has somekind of UI whether native or web or touch screen or something), which you can converse with to may be RAG stuff from some panel/tab and respond to questions.
I have seen way too many products suddenly becoming ProductX to now ProductX-AI by simply adding a RAG powered document conversation popup.
In fairness, as a company, you pretty much have to jump on the next bandwagon a lot of the time even if you have doubts whether it's there yet or is going to be a splash in the pan.
My story is that OpenStack didn't really work out but, if you were serious about the cloud thing, you sort of had to hop aboard even if, in general, the landscape ended up playing out differently with containers.
There is a big upside potential for high growth companies taking advantage of technology trends.
Today, Google’s revenue is £263.66 Billion. This is nearly 300x the revenue Google generated in 2003 ($961.9 million). The company went public on August 19, 2004, at $85 per share, valuing the company at $23 billion. After the IPO, Google reported $1.47 billion in revenue for fiscal year 2003, with a profit of $105.6 million.
But let’s ask a different question: aside from re-allocating the economy’s marketing and advertising budget into Google (from, presumably, local newspapers and TV before Google existed) how much of that revenue comes from actual tangible new wealth creation?
To put some context on this, 78% of Google’s revenue is advertising. Overall US ad spending has been increasing at about 1.6% per year since 2001, with no obvious indication of an acceleration (beyond some bumps around 2007/8.) So is there actually a success story beyond market capture here? And if all we’re doing is concentrating existing business into new channels, is this something we should be excited about?
You simply pay more for a product to find you. The more overpriced ones find you first. Googles business model is to make it as hard as possible for products to find you while simultaneously pretending to be the go-to place for precisely the opposite. A truly magical accomplishment.
Google created android - the most popular OS. Sure maybe samsung or nokia would be used instead but definitely the helped expend ad business with android. Same like Meta/ByteDance expanded Ad business with Intagram/Tiktok. Even if ad spending grew 1.6% per year it's not sure if it grew as much if android didn't exist. Also need to take into account probably reduced cost of advertising - this product just got cheaper. That the ad market grew 50% in 25 years doesn't mean we have only 50% ads served same like 50% grow (in $) in smartphone market doesn't have to mean you have only 50% more smartphones if they got cheaper.
This is an interesting topic and I’m not sure it has an answer. In 1995 advertising was really spray and pray. Testing ads was a really difficult proposition so we saw things like bearer coupons (mention this ad or bring in this coupon for 20% off!). The dominant advice out of radio was to play an ad constantly. That advice worked well for traditional media but not so well for advertisers. That model worked so well for advertisers that thirty years later, people around my age in my city can all sing the same five advertising jingles.
Google provided a toolkit to test ads and figure out which are most effective. Now the other side of that argument is that in industry, a massive of percentage of qualified people still spray and pray. The advertising industry as a whole is far from data driven.
At one point, there was an argument this was good for the planet. My newspapers are much thinner than they were 30 years ago when I could collect a metre of newsprint a month if I subscribed to the Globe and Mail plus a local. But I don’t think anyone can claim now that data centres are environmental miracles. This has also decimated local journalism to such a point that people are less aware of environmental catastrophes in their own relative backyards.
It’s possible the net effect was positive and advertising is more efficient. It’s more accurate to say advertisers have a toolkit to analyze effectiveness but many don’t or aren’t capable.
Edit - I’m going to give a very specific example of a radio jingle. If anyone is around forty or older and from a major city in Saskatchewan, they will be able to finish this.
“I said no no no no don’t pay anymore, no GST and no money down.”
Is that because of innovation? Or is that because of Google’s antitrust activity the US government is currently busting? Safari default search engine deal, etc.
"Antitrust" is just lawyer-talk for winning strategies that we later arbitrarily decide is not good for capitalism.
They weren't a bunch of gremlins in a cave conspiring to commit "anti-trust violations" in 2005. They were smart as hell and invested in the right areas.
Microsoft would get hit with the same anti-trust Google is being hit with if Bing and Windows Phone were successful - they're getting away with it because they're terrible.
Google search, google maps, gmail, YouTube, and Chrome have all been good functional products for over a decade. I genuinely don’t know what they’ve been doing since then other than milking us and getting new customers. Maybe 10% of this growth leads to a real improvement in human lives.
I agree with the article in faith, but i think they've gotten the cause wrong.
The problem is, scaling was ALWAYS the hard part. at a certain level, you don't have to worry about sharding and replicating databases, moving over to NoSQL, async race conditions, etc. etc. Why bet the house on one business idea, when you can have 10 "Micro-SaaS's" that are all bootstrapped but might make 10-20k in MRR.
In the day and age where the average business person has like 20-30 subscriptions for random tools, emails, websites, marketing, email lists, automations, SaaS products, freelancers, etc. it very much lends itself to the micro model.
The 'VC' business model is starting to break down. Just by looking around youtube and Indie Hackers, most of the successful businesses now adays are bootstrapped where the founder has some kind of community where they blog, youtube, have a patreon, X, etc.. They become the brand and they have no use for VC's. As soon as they launch a new app idea, they have 200K people on twitter, 150k people on youtube that will atleast give the app a look.
I agree that today's focus is more on integration than relying on a single "corporate" system. However, I believe a major issue with micro-SaaS in general is security. While even FAANG companies face security challenges, relying on many smaller SaaS providers introduce weak points into your system, and security is a challenging factor for small company budgets.
This isn't exclusively or particularly solved by VC-funded or giant tech companies - Dropbox once deployed a bad build that accepted any password, Apple accidentally had a blank root password, I'm sure there are many more embarrassing tales like this.
You are the exact people that i am trying to avoid with this model. I'm not trying to make big deals with big companies who can be impacted by security. The Micro-SaaS model requires that when i get a client asking me those kind of questions, i run from them and tell them my tools probably not for you. Any app that requires sensitive data transfering shouldn't be done on the micro-saas model.
Micro-Saas requires small, simple tools that may be low-hanging fruit. Sometimes they aren't micro-Saas's, but just random tools that make money for you by creating a glorified Open AI wrapper and a bunch of integrations. Honestly, alot of the tools I see that make money for people are made on Make or replit. No code even required but definitly not going after the "we need sensitive info or PII" market.
All payments just go through their respective provider so not really a risk there too.
The article somehow misses the mark on how VC firms actually make money apart from carry, its management fees. Thus, a16z raising a 20b fund on 3% management fee and 30% carry effectively guarantees them 600M even if the fund goes to zero and they have many such funds.
Sure, they would prefer to make money through carry, but the management fee is a nice downside protection.
Funds like A16Z can demand a high management fee because they have shown that they can deliver reasonable exits.
Most funds have management fees in the 1-2% range and a carry at around 20%. VC is a power curve, where a couple of large funds have an outsized impact.
And if a fund or VC (from associate to partner) cannot deliver, your career in the space is basically over.
IMO the best innovations tend to happen when someone has their hands bound and has no other choices but to figure out a new way or die. Now seems like an incredible time for new models and startup paradigms.
To me I think VC's figured out a way to market a very specific way to build companies and convinced a lot of people it's the only way for 20-ish years. Then there was this sort of shift to selling to enterprise, I think because B2C got harder and easy money was the goal. By then a lot of enterprise design makers were probably in the networks of the people selling. There's a meme about YC-of-late being mostly companies that sell shovels to each other.
But when you optimize for enterprise, I think you end up losing a lot of diversity of opinion in where the value comes from, which leads to top-heavy companies.
My main issue is that after the ZIRP era I don't believe the money is gone or unavailable. It just seems to be hoarded for some reason. There is astronomical wealth out there that could be used for trying new economic models that compete with the last generation of VCs. But it isn't happening.
Maybe the next era of VC decision makers, the ones who themselves were funded on big bets, just don't have the same appetite for risk? Or maybe the era of "developing your brand" has made them not want to share their success? I'm not sure but it's weird to me.
I guess it is time to do your homework and invest in real tangible products instead of trying to make a quick buck and get out before the company fails.
I can work for a year and produce something worth a few thousand dollars in profits.. meanwhile the financial grind generated that in a portion of a day. Those are the same dollars used to buy a house or save for retirement.
see Thomas Piketty .. this will get worse before it gets better
Startup wolf, wants to take 6 months off , build something, and pitch it non stop. Go to a random bar in SF and just sell random people on your idea. You only need one yes!
Worker wolf, sit down, work your 9-5. Drive a Prius.
A few of you have a 3rd wolf.
Bitter Wolf.
You started a company, got screwed over on some level, and now you're stuck thinking what if.
Readit News
The interesting thing, to me, is how speculative OpenAI's bet is.
IIRC it was 2019 when I tinkered with the versions of GPT 2.0 that had web interfaces, and they were interesting toys. Then I began using ChatGPT since its launch, which was around Dec 2022, and that was a profound paradigm shift. It showed real emergent behavior and it was capable of very interesting things.
2019 - 2022 was three years. No hype, no trillions of dollars invested, but tremendous progress.
Now, there has been progress in the part ~three years in synthetic benchmarks, but the feeling with ChatGPT 4.5 today is still the same as it was with GPT-3/GPT-4 in 2022. 4.5/o3 doesn't seem hugely more intelligent then 3.0 -- it hallucinates less, and it's capable of running web searches and doing directed research -- but it's no paradigm shift. If things keep progressing the way they're going, we'll get better interfaces and more tools, but it's far from clear that superintelligence (more-than-human insight, skill, and inventiveness,) is even possible with LLMs.
- Go down to your local Ray-Ban store.
- Ask to play with a pear of the Meta glasses.
- Activate "Live AI mode"
- Have a real time video conversation with an AI which can see what you see, translate between languages, read text, recognize objects, and interact with the real world.
Contrary to your (potentially misremembered?) history, nothing at all like this was possible in 2019. I remember finetuning an early GPT-2 (before they even released the 2B model!) on a large corpus of Star Wars novels and being impressed that it would mention "Luke" when I ran the produced model! Now I wear it on my head and read restaurant menus with it. Use it to find my Uber (what kind of car is that?) Today I am building my raised garden beds out back and reading the various soil amendments I purchased, talking about how much bloodmeal to put over the hugelkultur layer, having it do math, and generally having a pair of eyeballs. I'm blind. The amount of utility I get out of these things is ... very hard to overstate.
If this's "moribund," sign me up for more decay.
Maybe it's me having an extremely low imagination, but that stuff existed for a while in the shape of google lens and the various vision flavor of LLMs, and I must have used them.... 3 times in years, and not once did I think "Gosh I wish I could just ask a question aloud while walking in the street about this building and wait for the answer". It's either important enough that I want to see the wikipedia page straight from google maps and read the whole lot or not.
> an AI which can read text, recognize objects, and interact with the real world.
I can already do that pretty well with my eyeballs, and I don't need to worry about hallucinations, privacy, bad phone signal or my bad english accent. I get that is certainly an amazing tools for people with vision impairments, but that is not the market Meta/OpenAI are aiming for and forcefully trying to shove it into.
So yes, mayyybe if I am in a foreign country I could see a use but I usually want to get _away_ from technology on vacation. So I really don't see the point, but it seems that they believe I am the target audience?
Ironically, this typo is very likely a result of AI dictation making a mistake. There are a lot of common misspellings in English, like "their" and "there", but I've never seen a human confuse "pair" and "pear".
So yeah, there are cool demos you can do that you couldn't five years ago. But whether any of those cool demos actually translate into something useful in day-to-day life where the benefits outweigh the costs and risks is far from clear.
It’s the “boring” stuff that’s interesting: automating drudgery work, a better way to research, etc.
I’ve been predicting for years that glasses — whether AR or VR — are and will remain niche. I don’t think most people want them.
I somehow agree with the op, that I don’t think I’m much closer to hiring chatGPT for a real job in 2025 than I was in 2022, but also you that there has been meaningful progress. And in particular, products that are transformative for disabled people are usually big improvements to the status quo for abled people too (oxo good grips being the classic example—transformative for people with arthritis, and generally just better for everybody else)
How does this personal anecdote relate to the observation that (so-called "tech") "[v]enture capital is moribund except for AI. AI is moribund except OpenAI."
Is venture capital involved with the Meta Ray-Ban glasses that we are advised to try for free.
This certainly provides benefit to those with limited vision, which is great. But that is a very small segment of consumers. Besides those, how many other people do you know who are actually _using these glasses_ in the real world?
Google Glass came out 10 years ago.
Deleted Comment
Generative AI was a sort of paradigm shift, and can be developed into interesting tools that boost human productivity. But those things take time, sometimes decades to reach maturity.
That is not good for the get rich quick machine of Venture Capital and Hustle Culture, where quick exits require a bunch of bag holders.
You gotta have suckers, and for that Gen AI cannot be an "interesting technology" with good potential. It needs to be "the future", that will disrupt everything and everyone.
This is not entirely true, or at least the trend is not necessarily less hallucination. See section 3.3 in the OpenAI o3 and o4-mini System Card[1], which shows that o3 and o4-mini both hallucinate more than o1. See also [2] for more data on hallucinations.
[1]: https://openai.com/index/o3-o4-mini-system-card/
[2]: https://github.com/vectara/hallucination-leaderboard/
does it? An anecdote from yesterday: My wife was asked to do a lit review as an assignment for nursing school. Her professor sent her an example of papers on the topic, with a brief "relevance" summary for each. My wife asked me for help as she was frustrated she couldn't find any of the referenced papers online (she's not the most adept at technology and figured she was doing something wrong). I took one look at the email from her professor and could tell just by the formatting that it was LLM generated (which model, I don't know, but obviously a 2025 model). The professor didn't say anything about using an LLM, and my wife didn't suspect that might be the case.
My wife and I did some Google Scholar searches, and _every_ _single_ _one_ of the 5 papers cited did _not_ exist. In 2 of the cases, similar papers did exist, but with different authors or a different title that resembled the fake "citation". The three others did not exist in any form - there were other papers on the same subject, sure, but nothing closely resembling the "citations" either in terms of authorship or title.
People are expecting it to get exponentially better, but these kind of innovations are more a inversed power law.
The internet was adding a trillion dollars to the global economy by 2008 and the end of that rapid expansion, where AI is still sucking hundreds of billions a year into a black hole with no killer use cases that could possibly pay off its investment, much less begin adding trillions to the global economy.
And a decade before the web and internet explosion, PCs were similar, with a massive build out and immediate massive returns.
This excuse making for AI is getting old. It needs to put up or shut up because so far it's a joke compared to real advances like the PC and the Internet, all while being hyped by VC collecting companies as the arrival of a literal God.
We look at CPUs or the transmission of digital data and these seem to have improved exponentially but these are rather exceptions and are composed of multiple technologies at different stages. Like how we went from internet through phone lines, to dedicated copper lines for data, to optic fiber straight into to people's homes.
Eg: look how the efficiency of solar cells has progressed over the last 50 years
https://www.nrel.gov/pv/interactive-cell-efficiency
I think that's the key threshold all these companies have been running up against, and crossing it would be the paradigm shift we keep hearing about. But they've been trying for years, and haven't done it yet, and seem to be plateauing
And then in OpenAI's case specifically- this tech has become commoditized really quickly. They have several direct competitors, including a few that are open, and their only real moat is their brand and Sam's fundraising ability. Their UX is one of the best right now, but that isn't really a moat
What? MSFT is worth 3.25 trillion without scaling beyond any users. All these AI companies need is for everyone to pay them $100 a year for a subscription and they have more than justified their valuation. Same strategy Microsoft uses
I think you're misremembering how 3.0 worked. Granted, the slope from 2.0 to 3.0 was very steep, but a ton of progress has happened in the past few years.
Ie. better datasets.
Just recently I took a screenshot from a jira Burndown chart to write a description of the sprint progress for our stakeholders. Did it in one shot from a screenshot and got it right.
We had big progress in AI in last 2 years but have to take into account more than text token generation. We have image generation that is not only super realistic but you just text what you want to modify without learning complicated tools like ComfyUI.
We have text to speech and audio to audio that is not only very realistic and fluent with many languages but also can express emotions in speech.
We have video generation that is really more realistic every month and taking less computation.
There is big progress in 3d models generation. Speech to text is still getting improved and fast enough to run on phones reducing latency. Next frontier is how AI is applied for robotics. No to mention areas not sexy to end users but in application in healthcare.
Heck, I wrote myself my own personal radio moderator in a few hundred lines of shell, later rewritten in Python. As a simple MPD client. Watch out for a queued track which has albumart, and pass the track metadata + picture to the LLM. Send the result through a pretty natural sounding TTS, and queue the resulting sound file before the next track. Suddenly, I had a radio moderator that would narrate album art for me. It gave me a glimpse into a world that wouldn't have been possible before. And while the LLM is basically writing the script, the real magic comes from multimodal and great sounding TTS.
Much potential for really cool looking/sounding PoCs. However, what makes me worry is that there is not much progress on (to me) obvious shortcomings. For instance, OpenAI TTS really can't speak any numbers correctly. Digits maybe, but once you hand it something like "2025" the chance is high it will have pronounciation problems. In the first months, this felt like bad but temporarily acceptable. A year later, it feels like hilariously sad that nothing has been done to address such a simple yet important issue. You know that something bad is going on when you start to consider expanding numbers to written-out form before passing the message to the TTS. My girlfriend keeps joking that since LLMs, we now have computers that totally can not compute correctly. And she has a point. Sure, hand the LLM a tool to do calculations, and the situation improves somewhat. But it seems to be underlying, as shown by the problems of TTS.
Vision models have so many applications for me... However, some of them turn out to be actually unusable in practice. That becomes clear when you use a vision model to read the values off a blood pressure sensor. Take three photos, and you get three slightly different values. Not obviously made up stuff, but numbers that could be. 145/90, 147/93, 142/97. Well, the range might be clear, but actually, you can never be sure. Great for scene and art descriptions, since hallucinations almost fall through the cracks. But I would never use it to read any kind of data, neither OCR'd text nor, gasp, numbers! You can never know if you have been lied to. But still, some of the byproducts of LLMs feel like a real revolution. The moment you realize why whisper is named like that. When you test it on your laptop, and realize that it just transcribed the YouTube video you were rather silently running in the background. Some of this stuff feels like a big jump.
I disagree with 3.0, but perhaps that feels true for 4.0 or even 3.5 for some queries.
The reason is that when LLMs are asked questions whose answers can be interpolated or retrieved from their training data, they will likely use widely accepted human knowledge or patterns to compose their responses. (This is a simplification of how LLMs work, just to illustrate the key point here.) This knowledge has been refined and has evolved through decades of human experiments and experiences.
Domain experts of varying intelligence will likely come up with similar replies on these largely routine questions as well.
The difference shows up when you pose a query that demands deep reasoning or requires expertise in multiple fields. Then, frontier reasoning models like o3 can sometimes form creative solutions that are not textbook answers.
I strongly suspect that Reinforcement Learning with feedback from high-quality simulations or real environments will be key for these models' capabilities to surpass those of human experts.
Superhuman milestones, equivalent to those achieved by AlphaGo and AlphaZero between 2016 and 2018, might be reached in several fields over the coming years. This will likely happen first in fields with rapid feedback loops and highly accurate simulators, e.g. math problem solving (as opposed to novel mathematical research), coding (as opposed to product innovation).
Deleted Comment
It doesn't really matter how speculative the AGI bet is, their consumer AI business by itself is basically guaranteed to drown them in money. The only reason they're making losses at the moment is because they're choosing not to monetize their free tier users with ads, presumably since they don't need to make a profit and can prioritize growth.
But the moment they flip the advertising switch, their traffic will be both highly monetizable and ludicrously high margin.
1. People won’t ultimate go download ChatGPT.app or use a website — they’re going to be using the functionality through structured services in iOS and Android, and it’s necessarily going to be under control of the OS vendors for security/privacy reasons. This doesn’t mean Apple and Google own the LLMs — there will be consumer choice for antitrust reasons if nothing else — but the operating system has to be a conduit for access to your data, and also for unified user experience. Which means advertising will be limited.
2. Say it does go the way you think it will — what prevents a real non-profit, open source LLM from taking it away from the commercial players? There really is no moat (other than money, energy, data center space).
I think most people will continue to use Google and its Gemini-generated summaries at the top.
Deleted Comment
I wonder what “hugely more intelligent” would look like to you?
Also, in the (rather short) history of computing, what “hugely more X” has happened over the course of a couple of years?
They most certainly cannot
For context:
- GPT 2 was released in Feb 2019
- GPT 3 came out roughly 18 months later in 2020. It was a huge jump, but still not "usable" for many things.
- InstructGPT came out roughly 18 months later in early 2022, and was a huge advancement. This is RLHF's big moment.
- About 10 months later, ChatGPT is released at the end of 2022 as a "sibling" to InstructGPT. It's an "open research preview" at this point. This is around the time OpenAI starts referring to certain models as being in the "3.5 family"
- GPT-4 comes out in March 2023, so barely 2 years ago now. Huge jumps in performance, context window size, and it supports images. This is around the time ChatGPT hits 100 million users and is really becoming a reliable, widely adopted tool. This is also the same time that tools like Cursor are hitting the market, though they haven't exploded yet. Models are just now getting "good enough" for these kinds of applications
- GPT-4-Turbo comes out in November 2023, with way larger context windows and lower pricing.
- About 12 months ago, GPT-4o released, showing slightly increased performance on existing benchmarks over 4, but now with state-of-the-art audio capability support for something like 50 languages.
- 5 months ago, o1 releases. This is a big moment for scaling compute at test time, which is a major current research direction in ML. Shows huge improvements (something like 8x over 4) on some math/reasoning benchmarks. Within months, we have o3 and o4, which substantially improve these scores even further.
- In February of this year, we get 4.5, and then months later, the confusingly named 4.1, which shows improvements over 4o.
So to be clear, in 2019 we had an interesting research project that only a few people could tinker with.
18 months later, we had a better model that you could play with via an API, but was still a toy.
It takes more than two years to go from that to ChatGPT, and a few more months (nearly 3 years total) to get to the "useful" version of ChatGPT that really sets the world on fire. It took roughly 4 + 1/2 years to go from "novelty text generation" to "useful text generation".
In the 2 years since then, we've gotten multimodal models, a new class of reasoning models, baseline improvement across performance, and more. If anything, there is more fundamental research and wider variety of directions now (the kind of stuff that shifts paradigms) than before.
Dead Comment
VC money is in a constant state of FOMO, this is nothing new. Companies dress up as AI, or web3 or web2, or fintech or what ever to more easily attract capital. If 57.9% of dollars went to AI startups this year is not because everything is AI, I would bet 25% are just companies that tacked AI on to an unrelated business model and its skewing the statistics. I'll promise you that 10 years from now 57.9% of VC funding is going to be in some other buzzword and its not going to be AI.
When you need a dash of convincing b******t, they are excellent generators.
I have seen way too many products suddenly becoming ProductX to now ProductX-AI by simply adding a RAG powered document conversation popup.
Dead Comment
My story is that OpenStack didn't really work out but, if you were serious about the cloud thing, you sort of had to hop aboard even if, in general, the landscape ended up playing out differently with containers.
Broadcom is to thank for that.
There is no way both of those are even remotely true.
There is a big upside potential for high growth companies taking advantage of technology trends.
Today, Google’s revenue is £263.66 Billion. This is nearly 300x the revenue Google generated in 2003 ($961.9 million). The company went public on August 19, 2004, at $85 per share, valuing the company at $23 billion. After the IPO, Google reported $1.47 billion in revenue for fiscal year 2003, with a profit of $105.6 million.
To put some context on this, 78% of Google’s revenue is advertising. Overall US ad spending has been increasing at about 1.6% per year since 2001, with no obvious indication of an acceleration (beyond some bumps around 2007/8.) So is there actually a success story beyond market capture here? And if all we’re doing is concentrating existing business into new channels, is this something we should be excited about?
Wealth creation?
Google provided a toolkit to test ads and figure out which are most effective. Now the other side of that argument is that in industry, a massive of percentage of qualified people still spray and pray. The advertising industry as a whole is far from data driven.
At one point, there was an argument this was good for the planet. My newspapers are much thinner than they were 30 years ago when I could collect a metre of newsprint a month if I subscribed to the Globe and Mail plus a local. But I don’t think anyone can claim now that data centres are environmental miracles. This has also decimated local journalism to such a point that people are less aware of environmental catastrophes in their own relative backyards.
It’s possible the net effect was positive and advertising is more efficient. It’s more accurate to say advertisers have a toolkit to analyze effectiveness but many don’t or aren’t capable.
Edit - I’m going to give a very specific example of a radio jingle. If anyone is around forty or older and from a major city in Saskatchewan, they will be able to finish this.
“I said no no no no don’t pay anymore, no GST and no money down.”
They weren't a bunch of gremlins in a cave conspiring to commit "anti-trust violations" in 2005. They were smart as hell and invested in the right areas.
Microsoft would get hit with the same anti-trust Google is being hit with if Bing and Windows Phone were successful - they're getting away with it because they're terrible.
Dead Comment
The problem is, scaling was ALWAYS the hard part. at a certain level, you don't have to worry about sharding and replicating databases, moving over to NoSQL, async race conditions, etc. etc. Why bet the house on one business idea, when you can have 10 "Micro-SaaS's" that are all bootstrapped but might make 10-20k in MRR.
In the day and age where the average business person has like 20-30 subscriptions for random tools, emails, websites, marketing, email lists, automations, SaaS products, freelancers, etc. it very much lends itself to the micro model.
The 'VC' business model is starting to break down. Just by looking around youtube and Indie Hackers, most of the successful businesses now adays are bootstrapped where the founder has some kind of community where they blog, youtube, have a patreon, X, etc.. They become the brand and they have no use for VC's. As soon as they launch a new app idea, they have 200K people on twitter, 150k people on youtube that will atleast give the app a look.
https://techcrunch.com/2011/06/20/dropbox-security-bug-made-...
https://www.macrumors.com/how-to/temporarily-fix-macos-high-...
You are the exact people that i am trying to avoid with this model. I'm not trying to make big deals with big companies who can be impacted by security. The Micro-SaaS model requires that when i get a client asking me those kind of questions, i run from them and tell them my tools probably not for you. Any app that requires sensitive data transfering shouldn't be done on the micro-saas model.
Micro-Saas requires small, simple tools that may be low-hanging fruit. Sometimes they aren't micro-Saas's, but just random tools that make money for you by creating a glorified Open AI wrapper and a bunch of integrations. Honestly, alot of the tools I see that make money for people are made on Make or replit. No code even required but definitly not going after the "we need sensitive info or PII" market.
All payments just go through their respective provider so not really a risk there too.
Sure, they would prefer to make money through carry, but the management fee is a nice downside protection.
Most funds have management fees in the 1-2% range and a carry at around 20%. VC is a power curve, where a couple of large funds have an outsized impact.
And if a fund or VC (from associate to partner) cannot deliver, your career in the space is basically over.
To me I think VC's figured out a way to market a very specific way to build companies and convinced a lot of people it's the only way for 20-ish years. Then there was this sort of shift to selling to enterprise, I think because B2C got harder and easy money was the goal. By then a lot of enterprise design makers were probably in the networks of the people selling. There's a meme about YC-of-late being mostly companies that sell shovels to each other.
But when you optimize for enterprise, I think you end up losing a lot of diversity of opinion in where the value comes from, which leads to top-heavy companies.
My main issue is that after the ZIRP era I don't believe the money is gone or unavailable. It just seems to be hoarded for some reason. There is astronomical wealth out there that could be used for trying new economic models that compete with the last generation of VCs. But it isn't happening.
Maybe the next era of VC decision makers, the ones who themselves were funded on big bets, just don't have the same appetite for risk? Or maybe the era of "developing your brand" has made them not want to share their success? I'm not sure but it's weird to me.
see Thomas Piketty .. this will get worse before it gets better
Can I apply for YC again and get my annual rejection? So I can cry upper middle class tears.
I really need a business partner to keep me focused on features people actually want.
But my main business friend is focused on much more important things ( raising a new family) now.
Thinking about what's more important right now, making some games I know will make no money.
Creating a B2B startup that will also make no money.
Why can't you do this yourself?
I'm not business minded...
Startup wolf, wants to take 6 months off , build something, and pitch it non stop. Go to a random bar in SF and just sell random people on your idea. You only need one yes!
Worker wolf, sit down, work your 9-5. Drive a Prius.
A few of you have a 3rd wolf.
Bitter Wolf. You started a company, got screwed over on some level, and now you're stuck thinking what if.