They are doing both. Distilling Mythos down to affordable models, so they can continue to fund the business. And training Mythos level models at the high-end, to expand the frontier.
He didn't. That's why the article exists. You have to do the science to see if it does.
He was asking the question - do we see gains across other tasks? The underlying question was: Is the additional attention given to this specific task creating a false impression of progress?
Either you've memorized the outline (or detailed component shapes) of, say, a horse, or you haven't. Memorizing the outline of a pelican isn't going to help you with the horse.
You could train a model to do something a bit different like a pencil sketch, or vector graphic sketch, of something given a photo of it, and expect that to be a generalized skill, but if you are asking the model to do it "from memory" then memorizing a pelican is no substitute for not having memorized a horse.
I guess if you start an absurdly high number that makes no sense, then somehow group think only brings it down a bit and it’s still better than aiming for “true” valuation?
I have now had a chance to use it, and it never interrupted me. Weird call for the video to include that. Works very well, and can even mash up languages.
It’s not the same service. Google’s vision OCR is pure text extraction, not layout. Pretty sure Google’s doc AI services that can identify headers vs body text is $10 per 1k pages.
That’s true, though worth beating a dead horse to say that traditional OCR won’t hallucinate sentences, perform unwanted translation, or change the meaning of whole paragraphs to something more “appropriate”.
Not an easy thing to gauge. X/Twitter stopped publishing MAUs after the acquisition.
External estimates vary, some point to growth, some to stagnation. We know that revenue suffered. LOTS of partisan and emotional opinions for either.
Google trends does paint a bleak picture for X but I am also questioning how much google itself can gauge that after LLMs exploded in popularity.
Anecdotally I did notice that references and embeds to X are way less prominent and common than before. My usual news used to be filled with them. My consumption of the platform also plummeted after not being able to read threads when not logged, its much harder for me to get drawn into it.
Still without data, I would be surprised if the changes to verification and logged out access did not massively hurt new adoption. With tiktoks prevalence amongst a new generation I would bet its a matter of time before X gets grouped to tumbler/facebook, not dead, but way past its peak and cultural relevance. If that has not already happened.
Anecdata: Mastodon now has witty, un-earnest comments semi-regularly. And I have learned of maybe ten percent of breaking news things in the last three months on Mastodon via links that weren’t just to twitter.
Impact in the zeitgeist. Twitter was everywhere. Regularly referenced on live news television statins. Was considered a major news source. Shows had segments dedicated to it. Everyone and every brand had a Twitter. That is not the case today. Not by a long shot.
> Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor. This is called recursive self-improvement.
I doubt this is representative of real world usage. There is a difference between a few turns on a web chatbot, vs many-turn cli usage on a real project.
reply