thaanpaa

From this article alone it should be obvious that the long-repeated myth about frontier inference being subsidized is just that: a myth.

It's training and tuning new models that cost them money. And that's why they are trying to scare people into thinking AI will kill us all in the next 5 years; they want bans on the open-weight models because they know they won't be able to compete with them.

And yet enterprises will be the first to move to on-prem LLMs as soon as they become feasible. The next generation of TPU chips already promises 5x efficiency and enough RAM to run a 1TB+ model, and Kimi K3 is about as good as Fable for a lot of tasks, so we'll be there much sooner than anyone anticipated.

That's not a consequence of an LLM. It's a consequence of the training data. In fact, I would argue that the latest models aren't nearly as sensitive to the tone of input anymore. It's an issue that has been addressed by better curating training data.

You can just skip the clearly marked ad bit and read the rest of it. Jason Pargin makes a lot of good content on different channels, and throws in a book plug here and there. Doesn't diminish the quality of his writing in any way.

Me. I like the idea behind the product very much. I feel like my AI-saturated brain needs something like this very much at the moment.

I'm purely guessing, but they may be following the "early adopters pay premium price" model. Which means that there'd be a price drop in the fall, probably accompanied by a new model so that whoever bought the initial one stays happy.

Probably limits the number of intermediate tokens one way or the other. Almost certainly the impact on the result is close to zero.

Well, the fun part is that the algorithms themselves are deterministic. They are just so afraid of model distillation that they force some randomness on top (and now hide thinking). Arguably for coding, you'd probably want temperature=0, and any variation would be dependent on token input alone.

Voting for right-wing politicians repeatedly. You know, tax cuts for the wealthy, education cuts in fear of national debt, and all that jazz.

Transforming Normalcy into Epic Success!

It took a seemingly ordinary sentence and turned it into a thrilling adventure of epic proportions! The result was nothing short of mind-blowing, leaving us all in awe of our collective genius. Together, we conquered the challenge with grace and finesse, showcasing the power of teamwork and collaboration. Let's continue to push boundaries and defy expectations in all that we do! #EpicSuccess #Teamwork #Collaboration #Innovation

I think it's already happening. I've been using Claude a lot since 3.5. For a while, I barely bothered to check what it was doing because the code it generated tended to be fairly straightforward and usually worked on the first try. But in the last few weeks, I've started reading everything line by line again, because now it's not only often incorrect but also written in a goofy, convoluted way. AI-generated code being constantly fed back to it in a feedback loop would explain a lot, I think.

I use AI a lot these days, and the funny thing is that it feels much faster, but looking back, new features seem to take roughly the same amount of time, and the number of bugs appears to be about equal. The main time-consuming issue seems to be finding the right balance between providing too little and too much information. Give the AI too little reference data, and it begins deleting what it thinks is unnecessary, or it simply starts overlooking things; give it too much, and it chokes. I do spend less time writing or copying and pasting boilerplate code, but in the end, double-checking and debugging AI-generated code tends to equal it out.

I wouldn't be surprised if the first-quarter drop was even larger. Constant reports of poor quality haven't gone anywhere, and traditional car manufacturers are catching up and taking over. Meanwhile, Musk's popularity is rapidly declining due to the way he acts in public to the point of being high as a kite on TV, and people are finally starting to realize that he spends the majority of his time just tooting his own horn and being an asshole to those who actually work at his companies. Tesla's days could very well be numbered.

This is clearly about censorship, as anyone living outside of the United States can see. It's actually pretty obvious that they're banning the one platform where US oligarchs have no say over what goes viral and what doesn't.