theptip

21 ms·

421.
by theptip 1y ago
> AI can do anything a human can do - but better, faster and much, much cheaper. Should be pretty clear that this is a different proposition to the historical trend of 2% GDP growth. Mass unemployment is pretty hard for society to cope w
422.
by theptip 1y ago
AGI as defined these days is typically “can perform at competent human level on all knowledge work tasks” so somewhat tautologically it does threaten to substitute for all these jobs. It’s a good thing to keep in mind that plumbers are a th
423.
by theptip 1y ago
This is 1m nodes, you typically run tens or hundreds of pods per node, each with one or more containers. So more like 100m+ functions if I follow the Erlang analogy correctly?
424.
by theptip 1y ago
What I mean here is that this is certainly not what Dwarkesh would claim. It’s a ludicrous strawman position. Dwarkesh is AGI-pilled and would base his assumption of a world model on much more impressive feats than mere language understandi
425.
by theptip 1y ago
Fully autonomous agents are marketing fluff right now, but there is like $10T of TAM from promoting most knowledge workers to a manager and automating the boring 80% of their work, and this doesn’t require this full autonomy. Karpathy’s def
426.
by theptip 1y ago
Zvi made this point the other day, and then this counterpoint which I agree with more - if you think AGI is soon but you need to keep up the exponential datacenter growth for 2-3 years (or whatever “around the corner” means for the company
427.
by theptip 1y ago
> LLMs seem to udnerstand language therefore they've trained a model of the world. This isn’t the claim, obviously. LLMs seem to understand a lot more than just language. If you’ve worked with one for hundreds of hours actually exer
428.
by theptip 1y ago
> Secondly, if you re-analyzed the same malicious script over and over again it would eventually pass inspection, and it only needs to pass once You’d need some probabilistic signal rather than a binary one. Eg if some user with zero rep
429.
by theptip 1y ago
Is there a market for a distributed audit infra with attestations? If I can have ChatGPT audit a file (content hash) with a known-good prompt, and then share the link as proof of the full conversation, would this be useful evidence to de-ri
430.
by theptip 1y ago
No, that’s clearly not a goal of this project. This is a learning tool. If you want a local model you are almost certainly better using something trained on far more compute. (Deepseek, Qwen, etc)
431.
by theptip 1y ago
“Social Engineering”
432.
by theptip 1y ago
> some of the ones I've played with come from telling Claude "Here's my copy of programming book. Please read the book and pull out reusable skills that weren't obvious to you before you started reading This is actual
433.
by theptip 1y ago
It’s a useful intuition pump for value investment (check out Benjamin Graham), it pre-dates meme trading. Due to loss aversion humans are irrational about negative price movements, so you need some cognitive strategy to counteract. Here’s o
434.
by theptip 1y ago
Pre-training operates on a significant fraction of the entire internet. It’s simply not possible.
435.
by theptip 1y ago
> you don't need that plan because it's just an arbitrary block of pretty fungible work This is an over-generalization. If you are a team of 5 selling a B2C product, sure. But for B2B you tend to need to commit to a roadmap eve
436.
by theptip 1y ago
Honestly - consumers couldn’t care less, if the product experience lets them get things done. And especially if there is a free version. Google, Amazon, etc are some of the most popular brands in history, while they were doing much more egr
437.
by theptip 1y ago
The JetBrains AI integration is quite nice, but it’s a bummer that you need a dedicated subscription to an additional JetBrains product. I already pay for Claude.
438.
by theptip 1y ago
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com... We are still at 7mo doubling time on METR task duration. If anything the rate is increasing if you bias to more recent measurements.
439.
by theptip 1y ago
Yeah, it’s a very sensible feature to add. Obviously lots of users are making “please recommend a <product>” queries, it’s a usecase where deep research really shines. So if your users are spending a lot of time on a tasks, why not ma
440.
by theptip 1y ago
I thought OpenAI campaigned real hard against this one?
441.
by theptip 1y ago
I don’t think this is true. See https://arxiv.org/html/2211.04325v2 for example.
442.
by theptip 1y ago
The human brain has many systems that adapt on multiple time-frames which could loosely be called “memory”. But here I’m specifically interested in real-time updates to medium/long term memory, and the episodic/consciously access
443.
by theptip 1y ago
Agreed, it surely does require a world-model. I disagree that generic LLMs plus CoT/reasoning/tool calling (ie the current stack) cannot in principle implement a world model. I believe LLMs are doing some sort of world modeling an
444.
by theptip 1y ago
Sorry, to be clear I was making the stronger claim: I would consider it very foolish to use pure induction to bet on _exponential_ growth stopping within 1 year. I think you can easily find plenty of other long-lasting exponential curves. A
445.
by theptip 1y ago
> Where are they going to get more data? This is a great question, but note that folks were freaking out about this a year or so ago and we seem to be doing fine. We seem to be making progress with some combination of synthetic training
446.
by theptip 1y ago
> Why predict that the (absolute) growth rate is going to keep accelerating past exactly now? By following this logic you should have predicted Moore’s law would halt every year for the last five decades. I hope you see why this is a fla
447.
by theptip 1y ago
Type hints / gradual typing is crucially different from full static typing though. It’s valid to say “you don’t need types for a script” and “you want types for a multi-million LOC codebase”.
448.
by theptip 1y ago
Right, this was my thought. Can’t you just use a typing.Protocol on __getitem__ here? https://typing.python.org/en/latest/spec/protocol.html Something like from typing import Protocol class Indexab
449.
by theptip 1y ago
> a lot of technologies turn out to be S-shaped, not purely exponential, because there are limiting factors. Yes of course it’s not going to increase exponentially forever. The point is, why predict that the growth rate is going to slow
450.
by theptip 1y ago
If you now have two ERs within driving range, you have the choice to go to the cheaper one if you are conscious and in a stable enough condition to reflect. This is the sort of thing people already think about in the US.

More ›