Lease note that you are now member of the Resistance
=> More informations about this toot | View the thread
When you tell them "mates the web frontend doesn't work properly" and they reply "whoa, we see all correctly boo!" and then you make some tests with different computers, tablets and smartphones and quite often things do not work, and finally you get an error message and you send them the relevant info and then.... crickets
=> More informations about this toot | View the thread
"Scrum" is an old Latin equivalent for "scrotum", right? So "scrum master" is actually...
=> More informations about this toot | View the thread
Just ended a phone call with someone who defines herself as a "feline assistant" because she is, more or less, an ethologist specialized in cats
=> More informations about this toot | View the thread
Being scolded by ladies at the local supermarket because I'm not using loyalty points "correctly" is my new top achievement against capitalism
=> More informations about this toot | View the thread
If Sam Altman is wrong: with DeepSeek you have LLMs with the same performance of OpenAI's ones but a fraction of the total costs.
If Sam Altman is right: with DeepSeek you have LLMs with the same performance of OpenAI's ones but a fraction of the total costs.
Disclaimer: I know that I'm extremely trivializing many tech things here.
=> More informations about this toot | View the thread
I don't know if it helps, but Nvidia published a blog post detailing how to run DeepSeek-R1 distilled models on AMD Ryzen AI processors and Radeon GPU.
https://community.amd.com/t5/ai/experience-the-deepseek-r1-distilled-reasoning-models-on-amd/ba-p/740593
=> More informations about this toot | View the thread
I've always loved Brutalist architecture, so now I'm bracing for it to become mainstream in the wrong way. Because of "The Brutalist", you know.
=> More informations about this toot | View the thread
The fun thing about "open source AI" is that most people don't agree on what the definition means and most of self-defined open source AI projects aren't actually that open.
=> More informations about this toot | View the thread
I have not enough time to procrastinate all the things I have to do
=> More informations about this toot | View the thread
It sounds so "analog vs digital" that my car doesn't start because some mice chewed through some cables connecting some engine sensors to the car's electronic brain.
Nature strikes back. Orcas sink boats and now mice sink my car.
=> More informations about this toot | View the thread
Let's go and open a nuclear resctor to look at us bathed in the magical blue light of its Cherenkov radiation
=> More informations about this toot | View the thread
It's ok to be offended by how Chinese AI is censored, but that's the finger and not the Moon.
"Well it probably proves that AI techbros and VCs are fooling people, companies and the society at large, but hey I can't ask it about Tienanmen". Sure you want to follow this vibe?
=> More informations about this toot | View the thread
You wake up and during the night almost everyone has become an expert in AI technologies and geopolitical scenarios
=> More informations about this toot | View the thread
It's 7pm. I have here a FreeBSD USB stick and an old ThinkPad. This is not good for a Monday evening.
=> More informations about this toot | View the thread
So I tried to explain "situational awareness" to a friend who always runs with noise canceling earphones and, well, I failed.
=> More informations about this toot | View the thread
There's no timeline like the bubble timeline
=> More informations about this toot | View the thread
Big American AI companies have used the same approach corporations have with anything else: throw more and more money, infrastructure and computing power to solve problems (any problem) with brute force. Costs and externalities don't matter because, hey, investors and marketing.
Then, of course it comes out some small company proving that reasoning and optimization are more effective than brute force. Even if DeepSeek is much worse that we see now, it's still a proof that techbros are gaslighting the market selling inefficient stuff at high prices just because they can. And there's too much money involved to officially say that the presumed Kings have always been naked.
Mind you: we didn't actually need a DeepSeek to understand the trick. Just have a look at open source AI models' performance compared to proprietary (and supposedly more expensively but much better trained) ones: less than 5% difference for the best "reasoning" models. The market was already hyperinflated. #ai
=> More informations about this toot | View the thread
Basically:
a) LLMs will become commodity, sooner or later
b) SLMs give more or less the same results as LLMs, for most applications and right now
(a)+(b)= techbro panic "use our AI useless gimmicks plz or investors will eat us alive!!!1!!"
=> More informations about this toot | View the thread
You'll be able to use the Copilot key even on your Linux cogitator. I hope not to use the shiniy new Clippy AI... er, Copilot. But however, do as you please.
https://git.kernel.org/pub/scm/linux/kernel/git/dtor/input.git/commit/?id=907bc9268a5a9f823ffa751957a5c1dd59f83f42
=> More informations about this toot | View the thread
=> This profile with reblog | Go to bitzero@corteximplant.net account This content has been proxied by September (3851b).Proxy Information
text/gemini