TechTakes

smoke GPUs every day

I really want to see a Harari takedown.

It took a full eleven paragraphs before the article even mentions AI. Before that, it was a bunch of stuff about how Wikipedia is conservative and Gen Z and Gen Alpha have no attention span. If the author has to bury the real point and attempt to force this particular rhetorical framing, I think the haters are winning. Well done everyone.

Have you considered paying good money for a human artist to draw it for you? :)

If famed bag holder SoftBank are starting to raise their eyebrows when asked about future investments, the jig is definitely up

Weirdly the media are reporting that they have made a profit on their investments but when you actually read the articles, they are saying that the magical imaginary money that their OpenAI shares are worth has gone up

"Quitting your job is not just fun, it's invigorating!"

there's no real definition of the term, but dark money group usually refers a group that helps its secret funders influence elections, rather than a lobbying group

I like this reply on Reddit:

I do my PhD in fair evaluation of ML algorithms, and I literally have enough work to go through until I die. So much mess, non-reproducible results, overfitting benchmarks, and worst of all this has become a norm. Lately, it took our team MONTHS to reproduce (or even just run) a bunch of methods to just embed inputs, not even train or finetune.

I see maybe a solution, or at least help, in closer research-business collaboration. Companies don't care about papers really, just to get methods that work and make money. Maxing out drug design benchmark is useless if the algorithm fails to produce anything usable in real-world lab. Anecdotally, I've seen much better and more fair results from PhDs and PhD students that work part-time in the industry as ML engineers or applied researchers.

This can go a good way (most of the field becomes a closed circle like parapsychology) or a bad way (people assume the results are true and apply them, like the social priming or Reinhart and Rogoff's economic paper with the Excel error).

I was getting excited to read this but seeing the word "hyperstition" used three times in the abstract put a bit of a damper on things hahah

AI Singularity Fantasies : Tracing Mythinformation from Erewhon to Spiritual Machines

That title is a banger

I regret to inform y'all that the target of the blog post is a rat, or at least rat-adjacent

https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/

This represents a first-of-its-kind case study of misaligned AI behavior in the wild, and raises serious concerns about currently deployed AI agents executing blackmail threats.

I think there’s a lot to say about the object level issue of how to deal with AI agents in open source projects, and the future of building in public at all.

One of the few benefits of AI is that nowadays some PR threads are very entertaining to read.

Ugh, I'm so fucking tired of this shit.

I can imagine that an LLM can find bugs. Bugs often follow common patterns, and if anything, an LLM is a pattern matcher, so if you let it run on the whole world of open source code out there, I'm sure it'll find some stuff, and some of it might be legit issues.

But static code analysis tools have been finding bugs for decades, too. And now that an AI slop machine does it, it's supposed to bring about dystopian sci-fi alien wars?

Why are people hyped about that?

(Also this poster makes wrong claims about every exploit being worth millions and such, but the rest of it is so much more ridiculous, it drowns out the wrongness of those claims.)

"a zero day is an unknown backdoor" this shows both that they are trying to explain things to absolute noobs, and that they themselves dont know what they are talking about, a zero dayvis just a vulnerability which was not know to the people maintaining system. A backdoor is quite something else.

Also fuzzers also found 'zero day backdoors' and they didnt end the world.

til that youtube now features "posts"

....sigh

but I do know that what's available now is just f*cking impressive - and it will only get better.

Another victim of the proof-by-dopamine-hit fallacy it seems.

It's telling that the example he brings is that Claude can do pretty much decently what he was about to buy a 100$ voice controlled app for. As someone who aspires to the art of making great software, it's so infuriating to see how non-techies were conditioned into accepting slopware by years of enshittification and price gouging. Who cares if the tech barely works right? So does most anything, right?

Yeah he is trying to build his own EA movement. He also wrote a book (which I have not read) which basically argues that people in general are good not evil actually. (Fair enough, but not relevant).

Im still trying to meet him and shake is hand, the resulting matter antimatter explosion will take out the country.