theptip

32 ms·

631.
by theptip 2y ago
I reject the Stochastic Parrot theory. The claim is more about comparative advantage; AI systems already exist that are superhuman on breadth of knowledge at undergrad understanding depth. So new science should be discoverable in fields whe
632.
by theptip 2y ago
Math seems to me like the hardest thing for LLMs to do. It requires going deep with high IQ symbol manipulation. The case for LLMs is currently where new discoveries can be made from interpolation or perhaps extrapolation between existing d
633.
by theptip 2y ago
It’s hard because there are multiple dimensions being upgraded at different cadences. Architecture, parameter count, etc.
634.
by theptip 2y ago
Two bets with the same expected value can have wildly different variance. The OpenAI bet is clearly that they build AGI and become the worlds first $10T company. At this valuation you only need to think it’s P > 0.1 for it to be a reason
635.
by theptip 2y ago
> prices would be reduced by up to 33% (from $0.45 blended rate to $0.30), but PG&E’s profit margins are only 11%. This is addressed right at the beginning of the article. The argument is not that PG&E is skimming off huge profit
636.
by theptip 2y ago
Seems like a mostly reasonable list of things to not let AI do without better safety evals. > AI that tries to infer people’s emotions at work or school I wonder how broadly this will be construed. For example if an agent uses CoT and th
637.
by theptip 2y ago
I think this is a slightly different case no? If the ad network is using a very high precision variable to soft-link anonymized accounts, then randomizing the values between apps should break that. Your analogy applies more to things like t
638.
by theptip 2y ago
> Why do they need to know my screen brightness, memory amount, current volume and if I'm wearing headphones? This is clearly adding entropy to de-anonymize users between apps, rather than to add specificity to ad bids.
639.
by theptip 2y ago
It seems fairly clear that free speech does not apply to government employees acting in their official duty. The order is concerned with government censoring private individuals.
640.
by theptip 2y ago
So, a historical question - did these terms such as “pregnant person” get introduced by administrative diktat or by gradual evolution? I’m curious about the provenance of such terminology. Administrators have elsewhere made changes like ren
641.
by theptip 2y ago
I think training demand is what you might predict would plummet. Inference demand might increase but you could easily believe that there’s substantial inelasticity currently.
642.
by theptip 2y ago
The bear case is something like “investors are going to call BS on multi-billion training DC investments”. That represents most of their short-run demand. Not sure what is supposed to happen to the inference demand but I guess that could be
643.
by theptip 2y ago
The plan with Worldcoin is to use these unique IDs to airdrop a UBI.
644.
by theptip 2y ago
The aider benchmarks that swyx posted below suggest o1 is still better than r1 (though an oom more expensive). Interestingly r1+sonnet (architect/editor) wins though. This suggests r1 is indeed better at reasoning but its coding is hol
645.
by theptip 2y ago
I’m pretty sure there is funding on both sides; NIH vs DARPA for example. I suspect that NIH funds a lot more here and I’m skeptical that DARPA funds labs in China for this. The experiments WIV were doing were specifically targeted at ident
646.
by theptip 2y ago
> imperfect state of AI Don’t forget to compare with the alternative. Caffeinated staffers pulling all-nighters to ship a massive document will also make errors. > a law should never be so complex That ship sailed decades ago and AI h
647.
by theptip 2y ago
One thing I’m hoping we see: as laws become more tech-driven, perhaps we can compile them from a (more) formal model of sorts. If we can open these up too, then we can get laws which express the cost functions more directly. A big problem I
648.
by theptip 2y ago
The theory is quite clear, if you know which strains could hop to humans then you can prioritize monitoring them, just like we monitor influenza types in animal populations now. The problem is just that P(avert catastrophe) is fairly low, a
649.
by theptip 2y ago
I think the parent is arguing that lab leak is plausible, even if it wasn’t certainly the cause. GoF is foolish if you think the lab leak was remotely plausible. Most folks had no idea about the sort of GoF being done, and the attitude of m
650.
by theptip 2y ago
This sounds like a bad interview IMO. But also, don’t be this interviewee. Solving a problem like this with the type system is setting off all kinds of “too clever for their own good” alarm bells. The goal of a coding interview is to showca
651.
by theptip 2y ago
Yeah, I think monorepos will be better for LLMs. Easier to refactor module boundaries as context grows or requirements change. But practices like stronger module boundaries, module docs, acceptance tests on internal dev-facing module APIs,
652.
by theptip 2y ago
The arbitrariness is a risk. But the general idea is that there is a strong inertial bias against firing “enough”, and this grows exponentially with company size. Doing a PIP is extremely annoying when you are already sure you want to fire
653.
by theptip 2y ago
Sure. You seem to think that LLMs will be unable to identify abstraction opportunities if the code is not identical; that’s not obvious to me. Indeed there are some good (but not certain) reasons to think LLMs will be better at broad-not-de
654.
by theptip 2y ago
I’m not sure why that matters here. Users want code that solves their business need. In general most don’t care about repeatability if someone else tries to solve their problem. The question that matters is: can businesses solve their probl
655.
by theptip 2y ago
I think this misses a crucial dynamic. It’s not “custom scaffolding vs. wait for better models”. There is also fine-tuning. In specialized domains you can’t just rely on OpenAI finding all the training data required to render your experts o
656.
by theptip 2y ago
Well for one thing it’s a lot easier to communicate expectations to consumers at CES. “Coming soon to auction at a price set by users, no idea what that will be though, good luck!” is much less compelling for consumers trying to plan their
657.
by theptip 2y ago
This is interesting stuff, and a great stepping stone. I think the excitement around true agents will come when the AI can author the workflow pipeline, so to speak, in response to a request. This is an area where terminology is in flux but
658.
by theptip 2y ago
This is a decision for the US market first and foremost. The lawsuits you mention are sadly irrelevant to the decision-making; again, if you are about to be forced to make this change by Trump, the results of some cost/benefit study wi
659.
by theptip 2y ago
This lines up well with my experience. I’ve tried coming at things from the IDE and chat side, and I think we need to merge tooling more to find the sweet spot. Claude is amazing at building small SPAs, and then you hit the context window c
660.
by theptip 2y ago
Also, Zuck appointed Dana White to the board: https://about.fb.com/news/2025/01/dana-white-john-elkann-cha... So they have also given a board seat to a friend of Trump. But yeah, I think you're right tha

More ›