There's the 'look at it! It just makes sense!' type. There's the 'we wanted to do this thing but you're already doing it' merge. There's the 'let's be a monopoly!' merge, and its sibling, 'you are getting in the way of my aspirations to be a monopoly' merge. There's the quiet merge to deal with debts, the making-it-formal-but-functionally-we-were-already-merged, the hey-that-collaboration-went-well merge, and many more.
I am still holding out hope for the Drumstick-TikTok Merger. We need more humor these days and i'd really love to see someone like Drumstick take it over lol
I think it is more of, holy shit, OpenAI is entering this space, and maybe Anthropic as well. What can we compete with:
- Name? No
- Technology? No because we rely on other LLMs
Do we have anything? No. We've taken in a lot of money from big-name investors, so let's see if we can turn ourselves into a complementary asset before people realize that LLMs are nowhere near what they are sold as.
I would have agreed up until a few weeks ago. ChatGPT search is getting better, but kind of superficial so I still preferred Perplexity. But the new Gemini Deep Research is waaay better than Perplexity at deeper Internet searches and I imagine only will continue to get better.
Perplexity is useful as a thin layer of product over a base model. As Sam Altman said, eventually all such startups will be steamrolled by companies that own the models.
Sam Altman is not a credible figure, and that quote was rubbish IMO. There’s no inherent reason that foundation model trainers (the dumb pipes of the AI era) will win RAG by default. Apps like Perplexity aren’t even really constrained by the strength of the model. The secret sauce is the information retrieval, where OpenAI has no special advantage. But Google sure does…
He might say that, but I think in fact the opposite is true. The models are getting commodified. The real value lies in distribution (access to customers) and consumer product skills.
I have been using / fanboi'ing Perplexity since January 2022; overall, I am disappointed in the direction their product has been heading. While it is still the first URL I visit when I want apprentice-level help, I don't think this will be true for much longer.
If anybody at perplexity wants my more-direct feedback (beyond what I've submitted via your platform's conversations), my postal is listed in /hn/bio (I do not use email, so if your platform eventually `requires` this it's an immediate disusership from me dawg).
Not OP but it used to be that if you wanted an LLM that would cite its sources - Perplexity was one of the only games in town that did a really good job combining an LLM with an active search engine.
It was also much better for posing questions that required the most up-to-date knowledge.
It starts with a search and reasoning from there. Cites the sources in the middle of the sentences so you can just click and verify, see the whole context.
It blows every "model only" service because it's 100x more accurate.
Because they summarize and cite sources or because their models were trained on copyrighted materials? Summarization and training should be tranformative, and the user questions add that element of novel purpose to the original materials that should make the output non derivative. But most of their responses are one time use, nobody is ever returning to them.
Or due to their power, they've already secretly been taken over by the US Gov't. That's not really a "big conspiracy theory" at this point. I was mocked by the left for years for saying that the Gov't was involved in Facebook censorship. Turns out I was right. The biggest battle our Gov't has to wage is the battle for hearts and minds, and the control of information, and so they're trying to get in as deeply rooted as possible with every big AI company.
I don't know if it would come with the deal, but Bytedance web crawler is known to be the one with top number of requests per day among AI crawlers (src: https://blog.cloudflare.com/declaring-your-aindependence-blo... ) I guess one of Perplexity challenges is to have their own web index and of course that starts with having a powerful crawler. Also having a powerful crawler is useful for capturing tokens to train models. If that technology comes with the deal, it makes perfect sense for Perplexity to acquire them.
Funnily enough the Cloudflare blog identifies Perplexity engaging in dodgy practices to avoid robots.txt denylists:
> Sadly, we’ve observed bot operators attempt to appear as though they are a real browser by using a spoofed user agent. We’ve monitored this activity over time, and we’re proud to say that our global machine learning model has always recognized this activity as a bot, even when operators lie about their user agent.
Lol, I had to report Facebook using the documented Facebook crawler UA, coming from Facebook ASN as a bot to them because they misclassified it. Don't expect too much from their global machine. I wonder if this case also included people manually reporting it...
Readit News
There's the 'look at it! It just makes sense!' type. There's the 'we wanted to do this thing but you're already doing it' merge. There's the 'let's be a monopoly!' merge, and its sibling, 'you are getting in the way of my aspirations to be a monopoly' merge. There's the quiet merge to deal with debts, the making-it-formal-but-functionally-we-were-already-merged, the hey-that-collaboration-went-well merge, and many more.
And then there's this merge.
Suddenly tens of millions of people are hearing about Perplexity who hadn't heard of it so far.
[1]:https://www.instagram.com/drumstick/reel/DE07qPUywUg
- Name? No
- Technology? No because we rely on other LLMs
Do we have anything? No. We've taken in a lot of money from big-name investors, so let's see if we can turn ourselves into a complementary asset before people realize that LLMs are nowhere near what they are sold as.
Deleted Comment
Google has platforms Google also purchased Reddit user content. Meta had platforms and user content.
$500M is nothing to sneeze at, but that is like 3 orders of magnitude less than TikTok’s value.
If anybody at perplexity wants my more-direct feedback (beyond what I've submitted via your platform's conversations), my postal is listed in /hn/bio (I do not use email, so if your platform eventually `requires` this it's an immediate disusership from me dawg).
It was also much better for posing questions that required the most up-to-date knowledge.
It blows every "model only" service because it's 100x more accurate.
I use it instead of Google for every search now.
> Sadly, we’ve observed bot operators attempt to appear as though they are a real browser by using a spoofed user agent. We’ve monitored this activity over time, and we’re proud to say that our global machine learning model has always recognized this activity as a bot, even when operators lie about their user agent.
Clearly not working too well.
[!] Updated TikTok message: https://bsky.app/profile/gloomchen.bsky.social/post/3lg2xe4t...