theptip

28 ms·

601.
by theptip 1y ago
It’s “AI helps unravel”, not “AI discovers”. And it’s newsworthy, as AI-assisted discoveries are not yet boringly well-known. I think it’s cool to see, and a good counterpoint to the “AI can’t do anything except generate slop” negativity th
602.
by theptip 2y ago
Atomization vs gestalt is mostly orthogonal to the point here. Taking an extreme example to illustrate the point, if you could image every atom in the body you’d have the opportunity to understand the whole-system dynamics, or try to isolat
603.
by theptip 2y ago
> The theory here is largely mechanical I’ve long felt that the reliance on population-statistics (RCT) rather than individual diagnosis highlights how little we really know about medicine. A mechanic wouldn’t try to fix a car based on a
604.
by theptip 2y ago
I really struggle to grok this perspective. The semantics of whether it’s the LLM or the human setting up the system that “take an action” are irrelevant. It’s perfectly clear to anyone that cares to look that we are in the process of const
605.
by theptip 2y ago
I think there is a niche for both. Local LLMs are orders of magnitude smaller, so you could imagine cloud bursting for the difficult/important work like generating character portraits. That said it’ll be 10-20x cheaper in a year at whi
606.
by theptip 2y ago
An obvious one is for video games, interactive fiction, that sort of thing. AI dungeon with visuals could be pretty interesting.
607.
by theptip 2y ago
Yes! AI for docs is one of the usecases I’m bullish on. There is a nice feedback loop where these docs will help LLMs to understand your code too. You can write a GH action to check if your code change / release changes the docs, so th
608.
by theptip 2y ago
If you try to implement an actual state machine (or even interlinked ones like TCP) in this style you will have a very bad time. The FSM model is restrictive but this makes it much easier to exhaustively cover and validate all state combina
609.
Defeating Prompt Injections by Design(arxiv.org)
2 points by theptip 2y ago|0 comments
610.
by theptip 2y ago
This is fun! I think this sort of tooling is going to be very fertile ground for hackers over the next few years. Large swathes of the stack is commoditized OSS plumbing, and hosted inference is already cheap and easy. There are obvious sec
611.
by theptip 2y ago
> it suggests a strategic downgrade of the Chrome Hadn’t thought of it this way, but if there is (say) a 50% chance of being forced to divest Chrome, then the EV on your investments in the future are substantially lower. Strategic downgr
612.
by theptip 2y ago
_log(f”My variable is {x + y}”) Reads to me a lot more fluently to me than _log(“My variable is {}”.format(x+y)) or _log(“My variable is {z}”.format(z=x+y)) It’s nothing too profound.
613.
by theptip 2y ago
The phrasing is very precise here, it’s the first TPU for _the age of inference_, which is a novel marketing term they have defined to refer to CoT and Deep Research.
614.
by theptip 2y ago
Index funds seem fine, having some skin in the game is a good idea.
615.
by theptip 2y ago
Is the AMD GPU stack reliable for running models like llama these days?
616.
by theptip 2y ago
Would love to see their growth rates, but Basecamp seems like a canonical example of where renting makes the least sense. They are bootstrapped and not hyperscaling, so they don’t need flexibility for rapid and unpredictable growth (VC back
617.
by theptip 2y ago
Seems weird to me. The strength of Django is the ORM and the ecosystem/idioms. Why would I break compatibility with the latter by taking a fork? I would love to see better admin (so many have tried to do this) but it’s unclear to me wh
618.
by theptip 2y ago
It’s pretty good, the interesting thing is when it fails it seems to often be able to reason about what went wrong. So when we get CoT scaffolding for this it’ll be incredibly competent.
619.
by theptip 2y ago
Nope, it can. Got it in two requests, https://chatgpt.com/share/67e41576-8840-8006-836b-f7358af494... for the prompts.
620.
by theptip 2y ago
Just read the release post, or any other official documentation. https://openai.com/index/hello-gpt-4o/ Plenty was written about this at the time.
621.
by theptip 2y ago
Yeah, this is the obvious objection to the doom. Someone has to pay to train the model that all the small ones distill from. Companies will have to detect and police distilling if they want to keep their moat. Maybe you have to have an ente
622.
by theptip 2y ago
LLMs seem clearly better for learning the basics, which IMO is what this paper is measuring? Infinitely patient interlocutor who you can ask for code reviews on sketchy/embarrassing WIP. Maybe as the quality bar for generated code rise
623.
by theptip 2y ago
I am extremely skeptical that many people are making an informed cost/benefit here. I would wager most users don’t even know about the license terms. It’s the same as infosec in general. Most people don’t know about the risks, and anyw
624.
by theptip 2y ago
Unsurprisingly, both the OP and this are oversimplified statements. The more nuanced point is to note that simply reducing the world to accounting equations omits all of the human detail. Morale is a meaningful thing, or if you prefer, know
625.
by theptip 2y ago
But we are not discussing Chromium. Google’s browser is Chrome, and that product has search exclusivity deals that have been deemed monopolistic. Google could divest the Chrome product and keep contributing to Chromium, but the value propos
626.
by theptip 2y ago
> the fact that I can't tell what they stand to gain from this makes me suspicious. Meta is one of the biggest contributors to FOSS in the world. (React, PyTorch, Llama, …). They stand to gain what every big company does, a communi
627.
by theptip 2y ago
I’ve heard this described as talking in “Neuralese”. It seems plausible that this will be the most dense language for model-internal dialog (or presumably inter-LLM dialog assuming they share the same weights). You will penalize this inasmu
628.
Grok Grok(thezvi.substack.com)
7 points by theptip 2y ago|0 comments
629.
by theptip 2y ago
> We’ve also improved the coding experience on Claude.ai. Our GitHub integration is now available on all Claude plans—enabling developers to connect their code repositories directly to Claude Would love to learn a bit more about how the
630.
by theptip 2y ago
This looks great. One of the glaring holes in the ecosystem is a good ORM like Django or ActiveRecord; Diesel is simply more barebones. This looks like it’s a candidate for capturing the ease of use that these offer.

More ›