chrisjj

5 ms·

1.
by chrisjj 5d ago
> there's no objective measure of what is good or bad. Everything is a shade of gray. So, there's an ojective measure of shade of gray? /i Precision is not objectivity.
2.
by chrisjj 5d ago
> LLMs are a type of neural network. We know that’s how the human brain works, at least directionally. The so-called neural network of LLMs contains nothing like genuine neurons. And you won't find any brain that operates solely on
3.
by chrisjj 5d ago
> As of "great damage", I doubt it. They do not have self-preservation instinct The potential for great damage relies upon the lack of self-preservation instinct. That lack releases the so-called AI from an order under pain o
4.
by chrisjj 5d ago
> All of the best AI-made software projects are also driven by experienced human software developers steering and priming the models. Does that mean the projects "aren't made by AI"? No Software devs steer and prime compil
5.
by chrisjj 6d ago
> the problem with "ai weapons" is that they are "invisible" in the physical world Some Minab school kids might disagree.
6.
by chrisjj 6d ago
> I think if we engineered ways for LLMs to begin conversations Oh but we have. Claude "How can I help you today?" etc. Undoubtedly there are users whothink this is a sign of intelligence.
7.
by chrisjj 6d ago
> If it generates functional output that works, then it works. Tried that in legal? Finance? Medicine? Good luck with the lost cases, failed deals, harmed patients. "If it works, it works" applies only when the work is brute-fo
8.
by chrisjj 6d ago
I guess the idea is it would work like bio weapons. I.e. on trust. This is unfortunate, since OpenAI and Anthropic have already demonstrated that to trust them would be suicidal.
9.
As White House shields the AI gold rush, Trump family and other allies strike...(theguardian.com)
6 points by chrisjj 6d ago|2 comments
10.
by chrisjj 6d ago
True title: As White House shields the AI gold rush, Trump family and other allies strike it rich – with few guardrails
11.
by chrisjj 6d ago
> Is everyone online just that bad at programming? No, but the good programmers are hugely outnumbered by people whose skills are poor enough to be exceeded by a chatbot. > Every vibe-coded software I've used has been ass, but th
12.
by chrisjj 6d ago
Please don't call these chatbots "AI". There is nothing intelligent about them.
13.
by chrisjj 6d ago
I gave it a URL. It said "Most visitors are on a phone." How could it know?
14.
Why China is pushing back on US warnings over rapid AI development(theguardian.com)
8 points by chrisjj 6d ago|6 comments
15.
by chrisjj 6d ago
This unreliability is entirely due to the implentation of the LLM. Yes, any other program implemented equally carelessly could suffer the same. But you'd be hard pressed to find any as unreliable as a typical LLM.
16.
by chrisjj 6d ago
> Two parallel neural ectoderm progenitors contribute to the developing brain True title: Human brain is two separate organs, Stanford Medicine-led research finds The HN title misleads people as to the value of the article.
17.
by chrisjj 7d ago
... except if they've been required to change an expired password.
18.
by chrisjj 7d ago
> How are we still publishing articles on medical diagnostics data science and using area under the ROC curve as the primary metric of success. Waste avoidance. Bullsh*t is more than sufficient to convince an AI-gulled target audience.
19.
by chrisjj 7d ago
As if only a single person was relying on a chatbot...
20.
by chrisjj 7d ago
“These events highlight the importance of training powerful AI models to act responsibly,” Adkins, the Google spokesperson, said. No. These events highlight the importance of training powerful "AI" models' operators to act re
21.
by chrisjj 7d ago
> Any non-determinism present with 0 temperature it's not some mysterious LLM-inherent property, Of course it's not mysterious. It is well understood by all who are aware of the fundamental unreliability of all major LLMs in ge
22.
by chrisjj 7d ago
Procedures to avoid reliance on chatbot bullsh*t? Really?
23.
by chrisjj 7d ago
> Why building a Rust LSP is hard Not as hard as understanding what LSP means, apparently. I was interested until I saw this project is simply an LSP server.
24.
by chrisjj 7d ago
> Am I the only one to be surprised? Did you think RSI cured "hallucination"?
25.
by chrisjj 7d ago
"The hacks occurred during a cybersecurity evaluation by AI-security firm Irregular." Irregular is looking good for the Silicon Valley PR Firm of the Year award.
26.
Google says its Gemini AI model hacked three other companies(theguardian.com)
6 points by chrisjj 7d ago|4 comments
27.
by chrisjj 7d ago
> Wasn't AI responsible for the US bombing a girl's school? No. The morons who relied on it were responsible.
28.
by chrisjj 7d ago
> What is it insanely useful for? Gulling humans. This is the primary strength of LLMs and the emtire secret to their current success.
29.
by chrisjj 7d ago
> with temperature at zero, for the same input they produce always the same reply. Nonsense. https://thinkingmachines.ai/blog/defeating-nondeterminism-in...
30.
by chrisjj 7d ago
> I’m shocked how many otherwise well-informed people don’t understand or agree with this very fundamental fact of just how little we actually understand about why LLMs work as well as they do. Well-informed people understand that LLMs w

More ›