Amekedl

5 ms·

1.
by Amekedl 4d ago
Agreed; and it really is not that deep. Realistically; anyone paying for llm access (anthropic, openai, gemini), is getting their access, and a service provided billed by tokens, subscription, whatever. All the efficiency gains, which publi
2.
by Amekedl 13d ago
anyone read footnote 7 for a bench, where it says something about musical quartet 2023?? idk man until I test it, I won't get excited.
3.
by Amekedl 15d ago
everybody do be cooking with water. Chinese Labs provided pretty good, primarily cost-reducing techniques, like the sparse attention patterns recently. I'd bet OpenAI and Anthropic use their variants of those too, so they can get great
4.
by Amekedl 1mo ago
Being hard here, though: Purely dismissing it as not being worth of my time due to the example alone.
5.
by Amekedl 1mo ago
thinking output would be a good start
6.
by Amekedl 1mo ago
You got to know how to use the model+harness+prompt to achieve the results you want, but honestly for many projects and questions all the models already pump out their same best version of an answer. Sometimes it is really akin to a git clo
7.
by Amekedl 2mo ago
? How could Grokking be made workable? Mechanistic Interpretability, like using SAEs (EleutherAI did), also methods Heretic employs to decensor, surely you can see some "modding" does indeed achieve things. FYI, I prefer discussio
8.
by Amekedl 2mo ago
It's way overblown... The "anti LLM" or "no AI" clause is really more a "covering all ass" clause. If you want to improve Forgejo and have that AI PR; you better clean it up of all traces - resulting in a
9.
by Amekedl 2mo ago
No. 1. "Grokking" was shown on 4-digit modular arithmetic with a 1-layer transformer; this article extrapolates it to AGI and a $10B training run with exactly zero intermediate evidence. 2. The "small dataset" is 25 tril
10.
by Amekedl 3mo ago
The screenshot of ol' claude closed code with that ascii table tells it all: Vapor AIware. As if it really would work like that. The noise added by the verbosity alone is not taken care of enough, and this entire thing belongs on the g
11.
by Amekedl 3mo ago
I agree. This isn't what people struggle with to be honest. With the limitations, right now it is sloppy unfortunately... Yet cool to release it nontheless.
12.
by Amekedl 3mo ago
This entire steganographically marking feature is prolly vibecoded like everything regarding claude code. They can try sure, but classic cat and mice game dictates: thankfully the chinese WILL keep distilling.
13.
by Amekedl 3mo ago
Looks good. Won't ever buy a font though.
14.
by Amekedl 3mo ago
usage billing over the monthly plan when deepseek is over x25 cheaper?
15.
by Amekedl 3mo ago
Very, very early already with GPT-3. The simple fact that people will act on it and believe just because what they insinuated as a prompt and the answer being churned out on the screen looking somewhat readable. That alone was going to seed
16.
by Amekedl 4mo ago
DeepSeek rules. I'm using it to do stuff that's not too big in scope, because I still need to remain in charge. Even for this, western competitors have no chance, least Anthropic and OpenAI, plus Gemini also has gotten too expensi
17.
by Amekedl 4mo ago
Dev tools. The debugger is something for example that Microsoft ostensibly keeps to their own products, and how they totally slaughtered omnisharp. It killed my daily csharp vscode driver couple of years ago, only now catching back up somew
18.
by Amekedl 4mo ago
Thank you. Looking at the WebDAV in nginx, this is exactly what I searched for, wanted to read, and confirmed my suspicions ^^ But this one takes the cake truly... https://red.anthropic.com/2026/cvd/findings/A
19.
by Amekedl 4mo ago
Where is the link to the advisories then? :/
20.
by Amekedl 4mo ago
Agreed, also amazing citations in the parent comment ^^
21.
by Amekedl 4mo ago
I don't buy it. A lot of stuff this finds is also just simply wrong, benignly reported as true, despite upper/lower layers in the code burying the possibility of a vulnerability actually being exploited. It's a performance&#x
22.
by Amekedl 4mo ago
You are absolutely right! Kidding, but the analogy sits comfortably with me. I wonder though if this kind of behavior is potentially harmful, most likely less than drugs but nonetheless...
23.
by Amekedl 4mo ago
The future of open washing
24.
by Amekedl 5mo ago
I’d call it “open washing”, but it looks cool. Good luck with it
25.
by Amekedl 5mo ago
"The signal is not just stable. It also has high capacity." stopped reading right there also it's nothing that anybody using tails for example should have to worry about. Nothingburger.
26.
by Amekedl 6mo ago
depending how hard the "the brain is a muscle" saying applies, there is no way using LLMs/chatbot systems/AI is not going to deteriorate your brain immensely.
27.
by Amekedl 6mo ago
I worry about just becoming a product user. Some people push Claude and Claude Code at work, weights are closed, even Claude Code is completely closed and proprietary. If terms, which Anthropic controls, ever change, all work and time spent
28.
by Amekedl 8mo ago
This totally glosses over the debacle that was GPT-4.5 (which possibly was GPT-5 too, btw), and the claim that it'll ever outcompete humans also totally depends on whether these systems still require human "steering" or work
29.
by Amekedl 10mo ago
Care to elaborate regarding tauri? I figured it has matured into a fine electron alternative.
30.
by Amekedl 11mo ago
Regarding LLMs we're in a race to the bottom. Chinese models perform similarly with much higher efficiency; refer to kimi-k2 and plenty of others. ClopenAI is extremely overvalued, and AGI is not around the corner because among 20T+ to

More ›