jgalt212

5 ms·

1.
by jgalt212 6d ago
> This isn't even remotely true That's quite an extraordinary claim there which I doubt would hold up to an even cursory level of scrutiny. But if you yell it loud enough, maybe people will be afraid to challenge your assertion
2.
by jgalt212 6d ago
There is really is no excuse for Washington here. Even for the layperson+, it doesn't take much use of an LLM to figure out where they produce good and usable results and where it's of specious of value. + every layperson is an e
3.
by jgalt212 6d ago
is it easier for an LLM to figure out a CLI or a REST API?
4.
by jgalt212 6d ago
> I had Claude port CADO-NFS to run on GPUs. Then it orchestrated a fleet to run on scavenged idle capacity Is it easier to find unused GPUs than unused CPUs?
5.
by jgalt212 8d ago
> Isn’t the value in being able to verify the design before implementing? Fair enough, but to be truly safe you really need the whole pipeline connected. model -> validation -> emitted code.
6.
by jgalt212 8d ago
> Several of the new admissions, however, run counter to OpenAI’s fair use defense, particularly the rule’s requirement that use doesn’t substitute or harm the market for the original work. Sounds about right for the people and orgs invo
7.
by jgalt212 9d ago
> as very confusing for me to understand how you can trust the implementation unless you can compile TLA+ to C. Preach. I feel like I'm taking crazy pills every time someone claims TLA+ as the sine qua non of provably correct system
8.
by jgalt212 10d ago
> It takes less time for local model to take the first action on your task than it does for Claude to validate your login, put you into queue and start issuing the commands. This is only true if your local model is already resident in R
9.
by jgalt212 11d ago
> They’re hyping up their products as being able to destroy the human race then are somehow surprised at the level of anti-AI sentiment out there in the world. You hit the proverbial nail on the head there!
10.
by jgalt212 11d ago
Yes, these are very clever fuzzers. But the real problem is not the cleverness, but the willingness to spin up 100s or 1000s of subagents no questions asked. Must make those token numbers go up!
11.
by jgalt212 11d ago
Geographic Determinism https://www.litcharts.com/lit/guns-germs-and-steel/themes/ge...
12.
by jgalt212 12d ago
That's true, but then again, CNN is a mere husk of its former self. Something needs to change, but maybe it's too late for profit-driven news media.
13.
by jgalt212 12d ago
> All Oracle is doing is unpacking Nvidia servers and plugging them in. This is a commodity business But it's not if you can jump to the front of the GPU allocation line like Musk has done.
14.
by jgalt212 12d ago
RL works better / easier to implement with static typed languages.
15.
by jgalt212 14d ago
book pitch: On being sane in insane places: A tale from inside the marketing department of a frontier AI Lab.
16.
by jgalt212 14d ago
Of course, you're right. I was making social commentary / joke.
17.
by jgalt212 14d ago
Not including salaries, OpenAI spend $20MM to make $1MM. Very on-brand for a start-up.
18.
by jgalt212 14d ago
Disappointingly, no results for penis or vagina.
19.
by jgalt212 14d ago
Sounds a lot like crypto.
20.
by jgalt212 15d ago
I'd drop pandas if polars worked eamlessly with sklearn.
21.
by jgalt212 16d ago
5 years ago I could not afford top shelf coders. Now I cannot afford top shelf machines.
22.
by jgalt212 16d ago
I found this bit interesting. > Even so, an error in the theorem prover does not mean an error in the original result. For an incorrect result to slip through, the AI-generated proof would have to be wrong in a way that happens to exploi
23.
by jgalt212 16d ago
It's like Uber for thinking.
24.
by jgalt212 16d ago
True, but shops are exposing everything to Claude or Open AI. It's akin to outsourcing all manufacturing to subcontractors in China. That was cheaper, but extremely short-sighted. Now so many products have cheap Chinese knock-offs t
25.
by jgalt212 16d ago
And there's no proprietary IP in your company's app that you don't mind being sucked up into the training data?
26.
by jgalt212 17d ago
which part is wrong? > For any practical application, numerical solvers for Navier-Stokes already exist and do a good job. or > This proof is just checking the boxes for mathematicians.
27.
by jgalt212 17d ago
> Those problems can’t be formally verified with an automated theorem prover. It certainly seems like any problem that is amenable to reinforcement learning will be solved.
28.
by jgalt212 18d ago
I don't use agents. I like to code and AI with a REPL.
29.
by jgalt212 18d ago
A medium prompt = 1 million tokens?
30.
by jgalt212 19d ago
lawyers hope to trap / trick people that are smart enough to read, but not legally smart enough to understand enforceability.

More ›