pu_pe

13 ms·

121.
by pu_pe 6mo ago
> When you pay for a Claude subscription, what exactly were you promised? I was promised 5x or 20x the amount of resources that the free tier would offer. I implicitly expected the same quality too, not some watered-down version of the p
122.
by pu_pe 6mo ago
It seems every AI company eventually concludes that lobbying is required for them to operate. That suggests to me that they know they have no real moat.
123.
by pu_pe 6mo ago
So Anthropic is trying to save money on infrastructure, we all get it. However, it's not ok to degrade the performance your users have paid for. Last week the issue was that you reduced the default "effort" level, now the pro
124.
by pu_pe 6mo ago
This is partly why this talk about AI "solving science" should be taken with a grain of salt. Here the authors intentionally poisoned the publication record, but there are millions of papers out there that are also garbage, and it
125.
by pu_pe 6mo ago
I agree with the general insight here. Python is great for humans but once they are out of the loop it's no longer as useful. Having a compiler is more useful for LLMs indeed. However we are moving one step closer to complete inability
126.
by pu_pe 6mo ago
For me the reason would be to preserve traces of intentionality (ie what was the user trying to achieve with this commit?). These days a 10k LOC commit might be triggered by a 100-word user prompt, there is a lot more signal in reading the
127.
by pu_pe 6mo ago
I actually believe we need to rethink Git for modern needs. Saving prompts and sessions alongside commits could become the norm for example, or I could imagine having different flags for whether a contribution was created by a human or not.
128.
by pu_pe 7mo ago
Paper-thin, AI generated article that doesn't address its actual premise. We have no evidence that model collapse is happening at all.
129.
by pu_pe 7mo ago
It's crazy to me that this is not considered fraud. You sign up for a yearly plan under a given assumption of functionality, then they just change the terms to give you less than what they agreed to without compensating you in any way.
130.
by pu_pe 7mo ago
The English word for it is similar too (Mythomaniac)
131.
by pu_pe 7mo ago
I appreciated the writeup and your clarification. I wonder whether this was your first attempt to solve this issue with LLMs, and this was the time you finally felt they were good enough for the job. Did you try doing this switch earlier on
132.
by pu_pe 7mo ago
I really like the tool and how you designed the UI, well done! Very interesting use case and a slick interface.
133.
by pu_pe 7mo ago
Fascinating stuff. Any chance of using a sparse autoencoder or some other method to try to grasp what the model is actually doing in those middle layers? It would be quite cool to get a better sense of what type of input it is getting in th
134.
by pu_pe 7mo ago
> Like with any LLM project, the first 90% of the work was super smooth and barely needed my intervention. The last 10% was a slog. The author doesn't really describe which part was a slog, I thought autoresearch was supposed to be
135.
by pu_pe 7mo ago
How do you know that? We don't have access to the logs to know anything about its training, and it's impossible for it to have trained on every potential position in Go.
136.
by pu_pe 7mo ago
As another commenter mentioned, the point of the story from Borges is that a perfectly detailed map is rather useless, because you need abstraction (it's a repeating theme in some other stories from him like the Library of Babel, and F
137.
by pu_pe 7mo ago
The main thrust of the article is that codebases can grow too large to be manageable by LLMs. > It simply will not fit the context window, and README files are of limited use. I think many useful applications can be built without reachin
138.
by pu_pe 7mo ago
Rationalization and justification are relics of an era where having a convergent discourse was relevant. You had to make claims that the media class would spread and discuss because that's how our societies were organized back then. To
139.
by pu_pe 7mo ago
It's not that hard for you... the process you just described is unintelligible for 99% of the population I would say. And then you have to produce the content on top of that.
140.
by pu_pe 7mo ago
First thing I thought... honestly we should bring it back anyway
141.
by pu_pe 7mo ago
Nice piece, and rings true. I also think startups and smaller organizations will be able to capture better value out of AI because they simply don't have all those approval layers.
142.
by pu_pe 7mo ago
The original report is very well written [1], and has good actionable recommendations too (interoperability, portability, decentralization, creating new competition). [1] https://storage02.forbrukerradet.no/media/2026&#
143.
by pu_pe 7mo ago
From your description it sounds like they could offer SpaceX at a valuation of $10T and be able to still sell because every passive fund is automatically forced to buy it. But obviously they need to target a price range that will be more or
144.
by pu_pe 7mo ago
Benchmaxxing aside, if you are using those tools for programming on a regular basis it should be self-evident that they are improving. I find it very hard to believe that someone using LLMs today vs what was available one year ago (Claude C
145.
by pu_pe 7mo ago
Even if your goal is to go out and create value for others, your contribution is proportional to what everyone else can offer. If others with AI will deliver that value cheaper, or if what I am good at can be easily automated, it's get
146.
by pu_pe 7mo ago
If anyone can go in, take a GPL project like chardet and reimplement it using LLMs, then the current maintainer just saved everyone time by making their implementation publicly available. Our legal framework wasn't built for a situatio
147.
by pu_pe 7mo ago
OpenAI and Anthropic have ~40 billion yearly revenue already, and growing fast. I don't know if 10x that is the right amount to invest, but I'm pretty sure 0.1x of that is not going to cut it.
148.
by pu_pe 7mo ago
European AI investment plans are always so timid. This article mentions a couple of investments on the order of $1.5-2.5 billion USD that their government says "would position the UK as a world leader in AI". Hard to imagine you c
149.
by pu_pe 7mo ago
I meant trivial in the sense it's a solved problem, I'm sure it still costs a non-negligible amount of money to train it. See for example the chess transformer built by DeepMind a couple of years ago which I referred to in a sibli
150.
by pu_pe 7mo ago
There are chess engines based on transformers, even DeepMind released one [1]. It achieved ~2900 Elo. It does have peculiarities for example in the endgame that are likely derived from its architecture, though I think it definitely qualifie

More ›