benjiro29

7 ms·

1.
by benjiro29 6d ago
> I think we should also point out that H200/B200's are seriously overkill for I simply mention what came to mind ;) A quad 6000 with 96GB, can run this model at NVFP4. That is 60.000 Euro for the GPUs and lets be generous with
2.
by benjiro29 6d ago
it's not really flash anymore, imo. Flash is about speed ... Flash models are supposed to be fast, way faster then their big brothers that are "better" but way slower. Its just that up to now, getting more speed involved cu
3.
by benjiro29 12d ago
Really depended when you bought them .. 386 is like the point where people really started to buy PCs. My 386 was more to the end period, so the jump to 486 was very short. Do not forget there was a ton of 386's and 486s released also.
4.
by benjiro29 12d ago
Are you not better off with cheaper models like GLM 5.3 or GLM 5.3 Flash? As a lot of that work is repetitive and only needs a stronger model at later stages of cleanup, no?
5.
by benjiro29 12d ago
3. The problem is also that at a lot of code bases are not designed around LLMs and their token usage. If you have a monolete codebase, you need to clearly define in the prompt what modules are involved. And even then, your wasting tokens w
6.
by benjiro29 13d ago
My first PC 386 was in todays money easily $5000+ (basic 2d GPU + screen)... A lot of hardware in our family was handed down to my folks, because you lost so much on selling, that it was better to keep using them as they had less demands. 3
7.
by benjiro29 13d ago
What was their rationale for doing nothing? Carrot: Free vacations? Merch? Benefits ... Stick: Scared because taking actions against the Buy (aka you own nothing) is paramount to going in a fight with a entire industry.
8.
by benjiro29 15d ago
I hope that this also applies to the Subscription usage. As that can then stretch out Fable usage by a lot more.
9.
by benjiro29 19d ago
OpenCode Go is probably using quantized down DS4Flash. They outsourced to 3th party providers to keep the cost down, and being able to provide that $30 value (instead of the initial $60 > $15). We saw the same issue with GLM 5.2 when the
10.
by benjiro29 19d ago
You will notice that DMCA claims are often against smaller parties. You rarely see those DMCA claiming companies go after somebody like Microsoft because those companies can fight back. Its a system that mostly benefits large companies. Jus
11.
by benjiro29 19d ago
The problem with DMCA claims is that there are no consequences on misuse. It places all the work on the affected parties to prove their innocence. And suing the fake claim, is years of work and cost. This is why companies like Tracer.AI, ..
12.
by benjiro29 21d ago
Its not the end of programming, its the change from how we program. Do we still write code in Assembly? No, we moved over to a form of programming that allowed more people, to easier program. Did it mean that the assembly guys lost their jo
13.
by benjiro29 22d ago
From my experience: * OpenAI was ~$170 per week in value * Claude was around $500 per week in value. * Ollama Cloud is giving me about ~$160 per week in value. So right now (thing constantly change in the AI world), its not more generous th
14.
by benjiro29 23d ago
I think it really depends on what people expect from the models and how they "code". Sol in my eyes is powerful, but it over engineers so much, that its actually a liability. Where as Opus 5 is slightly under develops but you then
15.
by benjiro29 23d ago
*This is likely because of your thinking level. The difference between max and ultracode is primarily that the latter is max with a bunch of agents.* I do not know why people even use Max or Ultra levels of thinking effort. Most of the time
16.
by benjiro29 23d ago
It did not exactly help that we saw traffic to DS (over OpenCode) jump from around 1.6T tokens per day, to over 14T token in a matter of days. Nobody has the compute to deal with such increases. This keeps happening with every good new mode
17.
by benjiro29 23d ago
I spend way too much time in all the LLM related subs, to the point that i consider it unhealthy (inc claude/anthropic subs). Its in my opinion not wide spread at all and as today is literally the first time i ever hear anybody mention
18.
by benjiro29 25d ago
* Anthropic's Cyber Verification Program // Codex + gotTAC approved* Meanwhile the Chinese models are "go ham dude"... If it was not for capacity issues, Chinese models have a higher change to just dominate. > $2
19.
by benjiro29 25d ago
*Nobody is willing to help with Iran even on symbolic level* Most people do not realize how much of a shift that is. In the past there was always a ton of EU countries helping the US out, in whatever crap the US started. And a lot of Europe
20.
by benjiro29 26d ago
I am guessing its GLM 5.3 Air + Vision. A smaller then 250b model.
21.
by benjiro29 26d ago
Benchmark results: https://x.com/deepseek_ai/status/2090730032574631962 Vision + increase in benchmark results.
22.
by benjiro29 29d ago
Only on openrouter for some reason. Not OpenAI Subs/API.
23.
by benjiro29 29d ago
Seeing how many people with Max accounts on Codex are complaining, your in for a rude awakening. The days that Codex was the undisputed usage king, seem to have been reversed. Its been most most noticed after they did like 20 resets in a mo
24.
by benjiro29 1mo ago
Fixed ...
25.
by benjiro29 1mo ago
> That trust was what led to this incident where someone just walked into the airplane dressed as a maintainence worker and nobody stopped him. O, it can be even worse, when you realize its a 30 year old problem ... During the 1998 te
26.
by benjiro29 1mo ago
The problem with DS Flash/Pro is that they are extreme reasoning heavy and step heavy. Step = cache hit. Reasoning = output hit. So the impact on those price increases will be felt much stronger. I think that Flash is still a usable mo
27.
by benjiro29 1mo ago
Strange that i do not experience this. Its been great in my experience. But that may simple be because i switched from typing most of my prompts. To just dictating my prompts in a long and convoluted way and letting the LLM extra the inform
28.
by benjiro29 1mo ago
I thought it was impossible to downvote posts? User Posts can be downvoted but you need over 500 karma to have access to the downvote button. A Submission can not be downvoted.
29.
by benjiro29 1mo ago
Its funny because "Shall we get rid of the FAA and the FDA then?" ... matches exactly with the current situation of the FDA and Points 3 > 5 ... The systematic nerfing of the FDA and other organizations their power.
30.
by benjiro29 1mo ago
HBM requires stacking the chips. So they need to shave the layers, glue, stack more, shave again. They also require a substrate what is even more wafers. The issue is that a error in the stack means a lot of losses. In order to get high ban

More ›