@AlfieOiya@Ariszn12 The pharmacy. They’re always at the pharmacy and somehow their problem hasn’t been solved in their last 60 years of picking up prescriptions
@Walsh007Billy@WehrnerMagnus Did/do you smoke a lot of weed? Did you get addicted to your phone? If both no, then something might be wrong with you. If yes to one, you’re going through an addiction or recovery that very commonly fucks with your cognitive abilities but it can get better
@Bolmercl@thdxr@RhysSullivan Not as much anymore honestly. If it’s complex enough to hallucinate I just won’t use google tbh, if you keep the queries straightforward enough it’s fine now. But it wasn’t just months ago
@trydotworks@thdxr The cost of a few tokens they can charge the user is probably so much less than the internal infrastructure headache of HBM not being available to serve the cache if everyone did this a lot. They’d need to either offload more of the KV to more tiers or secure more memory, I guess
@michellechen@Cloudflare This was a very well written article, and it actually addresses KV cache penalties in far more depth than other routers. Nice job whoever thought of this and whoever wrote it up!
@userkrisg @olvrgln Nobody has a problem with CPU data centers. They use far less power, far less water, and don’t require specialized natural gas generators to power when they can’t get enough from or connected to the grid.
Your comparison is naive and uneducated. A CPU is not a GPU.
@johnspaetzel@GMMeyer@alvinsng Leetcode is still the only thing that matters, and I think we all know it provides zero applicable programming experience. It’s good for the pre-AI FAANG structure. Nothing has changed in how interns or juniors get hired despite AI, for the vast majority of companies.
@GMMeyer@alvinsng They’re also expected to do that BEFORE formally entering the workforce. The jobs for entry levels are the worst they’ve ever been. Some people would be fine doing that, but they literally don’t have the door open to make that call.
@LevyAntoine *some researchers and alignment employees
** “may” meaning non-negligible percent chance
*** hedges in positions across stocks/futures/etc are cumulatively worth trillions of dollars and the chance they succeed are likewise non-negligible but also not probable
@warbird__actual@TheStreetsofATX@esjesjesj And no “liberal” is going to tip to voting Paxton or abstaining, unless they’re retarded or a single-issue Transgenderism supporter in TEXAS.
Is it more likely this liberal exists or the moderate Christian exists, in Texas? This is Texas, not California or Oregon.
@DavidSKrueger I don’t know why they try to preface with that.
When has anyone ever led with “I’m not left or right, but…” and then delivered an objective and unifying statement? lol
@poordart@MenchOsint No. No no no. This cannot be true. The Zionist democrats (the good guys) have reliably assured me that Israel is not the problem, it’s just Netanyahu. You are wrong.
@maxwinga@coffeebreak_YT How do you plan to legislate this in the countries that matter? I don’t think the UK actually has any labs that work on AI that can compete, so no other country will care. Even Mistral, a borderline non-competitive lab, isn’t there. How will you work with US and China?
@0xWren@_lopopolo https://t.co/U1HBSnK55x
lmfao I guess workflows are not skills but some higher level markdown form of harness engineering, while all other “skills” are bad attempts to constrain the LLM which could otherwise be deterministic code.
Fascinating.
@monkeymktplace emphatic disagree. if you had deterministic processes you'd use the token machine to write code. putting determinism in text puts the agent in the box and wastes its potential
@0xWren@_lopopolo Yeah this guy might be retarded or simply looking for attention and we gave it to him.
In his other replies he seems to have some flawed notion that a skill is constrained to some validation-like deterministic process in markdown that could be better as deterministic code.