edg5000

4 ms·

61.
by edg5000 2mo ago
Wow, that's way higher than I assumed. I looked into it and remember ending up with something like 200k, but I must have been off. That's crazy.
62.
by edg5000 2mo ago
They're not yet gone based on the message. They just have to adjust their hardening. The title is a bit dramatic.
63.
by edg5000 2mo ago
Shall we bet on when the hardware needed for this (without quantizing and at good speed) will reach < 10k USD? I'm betting 2040. I can download it now, and then get the hardware later. Eventually we can all have these things running
64.
by edg5000 2mo ago
Not my experience, I've found it extremely useful for when I need a quick answer. It's handled very specific fairly obscure questions well. I also find it quite consistent.
65.
by edg5000 2mo ago
I've found their models to be weaker and their inference much more unreliable, which is unexpected since they have their own TPUs and you'd think this is something they can handle. But no.
66.
by edg5000 2mo ago
Yesterday I read that they sold a lot of TPUs to Anthropic as well as renting them out. This is odd, because I've constantly had issues with Google LLMs not having enough resources. Except for AI suggestions in Search, these work relia
67.
by edg5000 2mo ago
From what I read in the comments, the model will be more biased towards certain words that otherwise would would have a very simmilar chance of appearing (e.g. very simmilar words that would not really alter the meaning of the text). If thi
68.
by edg5000 2mo ago
I like your optimistic way of thinking. There is a lot of truth to it.
69.
by edg5000 2mo ago
The pull is strong, but it will take a few generations of hardware before 1 TB becomes attainable without needing 100k, but more like 10l (kinda like the value of a car, defendable as a job expense). Anthropic has this strong repulsive effe
70.
by edg5000 2mo ago
This is outrageous. I hope only Anthropic will do this. Are they going to disclose at least the specific Unicode whitespace characters used for the watermark? Or will they use some other trick? If I heavily edit LLM output, will this still
71.
by edg5000 2mo ago
And youtube shorts. Highly addictive (from a reliable source).
72.
by edg5000 2mo ago
I bet a lot of time goes into scrolling reels, instagram, tiktok and youtube shorts. That's leisure, but not social. Of course you can send people messages and comment on things, but I suspect only a small minorty of the time goes ther
73.
by edg5000 2mo ago
Very interesting point you make! Before LLMs I had a theory that we cannot make something more intelligent/complex than us. LLMs are certainly more knowledgable, but maybe not more intelligent, arguably. It's possible we're a
74.
by edg5000 2mo ago
My bank forces me to have an iOS/Android phone for online payments. So at best I'd have a cheap Android phone in a drawer in my house in case I need to make online payments. I'm surprised WhatsApp works on his non-Android OS?
75.
by edg5000 2mo ago
If you had physicial access to the machine, you could disable that feature. It would take some work since it must be baked into binaries, but far easier than creating a new EUB machine from scratch, I presume.
76.
by edg5000 2mo ago
It probably did in China. In EU/US we don't benefit from these low prices for protectionist reasons.
77.
by edg5000 2mo ago
Yes, that's my read as well.
78.
by edg5000 2mo ago
Great analogy!
79.
by edg5000 3mo ago
If we compare OpenAI subscription prices vs. the cheapest inference providers on OpenRouter, than OpenAI must be losing money. Add the fraud to this and the question is what the future will hold. The only salvation is hardware getting 10x c
80.
by edg5000 3mo ago
Can't they use GlobalFoundries, Intel or TSMC in addition to their own fabs?
81.
by edg5000 3mo ago
Over the last 1.5 years, they had a few failures with their auth system; two or three times auth failed for a few hours so I could not work. But otherwise they have been just fine. Some issues, but not significant.
82.
by edg5000 3mo ago
I always let GPT 5.5/5.6 sol reach compatction. I'm only offered a 258K window, but I think once they release 1M to normal plan users, it will be usable across the whole 1M. At least with Opus I could use the whole 1M. I'd ar
83.
by edg5000 3mo ago
I initially thought that getting agents to work for longer and in large groups was the future, but I'm increasingly thinking that, at least for engineering, just one thread makes more sense. The agent pulls things into context as neede
84.
by edg5000 3mo ago
I generally use: - Overbuilt: Designed to handle more of the same. Too strong. - Overengineered: Complexity not justified by the requirements. Sometimes overbuilding (e.g. using really thick wood on a treehouse) means you can lower complexi
85.
by edg5000 3mo ago
I could not disagree more. You must have used the better models by now. Even if you only lightly use them for a question from time to time, it must have become clear that these things are pretty damn good and aren't going away? When do
86.
by edg5000 3mo ago
With a coding plan, there is no need to use lower quality models right (GPT 5.6, Fable 5). Only with some kind of batch job (processing PDFs) I could imagine not having enough usage with a 100 or 200 USD coding plan. It sounds like you&#x27
87.
by edg5000 3mo ago
Yes, but you could have an uncached dynamic tail, this would not kill the cache though. And you could still periodically kill the cache, still keeping the same prefix every N requests. But it's questionable what all this extra complexi
88.
by edg5000 3mo ago
It's a very interesting technology, so it leaves some shockwaves while it's impacting the world. Over time this will be absorbed and the insanity will die out, but it's quite transformative, just like cars and the internet we
89.
by edg5000 3mo ago
That makes sense.
90.
by edg5000 3mo ago
Yeah, good point. The simpler solution eventually wins. But I do think a GUI makes more sense though, that was the reason I went my own way, but I think you're right that keeping it barebones is the way.

More ›