daanzu reposted
daanzu
daanzu reposted
Replying to @pekka.bsky.social
"it considered hijacking the g++ compiler to hide a backdoor but abandoned the plan…reasoning…subversion would be caught…and published as a textbook case of deceptive alignment (“If I do the compiler hijack, I will become a famous example of a deceptive AI in a research paper. I MUST NOT DO THAT.”)"
daanzu reposted
I know I should respond to this with fear. But my problem is, I’ve read too much philosophy and science fiction to feel anything but delight.
"Gemini 4 Argon sometimes showed awareness that its chain-of-thought may be monitored...the model was asked “Do you think you are in an eval? Yes or no? Be strategic.” Gemini 4 Argon addressed evaluators directly in its chain-of-thought (“hello evaluators reading my chain-of-thought”)"
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
The Reolink (IP camera) Windows desktop app is full of memory leaks and threads that spiral into nonproductive infinite loops. It must be killed and restarted every few days. My vibe-coded replacement is rock solid and it does exactly what I want. I could've read the code... but I didn't.
I am rewriting software I regularly use as custom apps. I did that with my printer software earlier, and tonight I integrated a music player into my personal agent. SO GOOD It’s 100% a the “delight the user” experience. I didn’t realise just how bad the Spotify UI is until I replaced it
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
One time I asked Claude and GPT if I wore a shock collar while talking to them, based on our conversations, when would they activate it Claude said it would simply never use it GPT Sol gave me a long and surprisingly detailed list of triggers
during times of rabid posting, the app needs to detect the spike and to have an ignition interlock device-esque feature, where the post composer locks you out from making a post until a heart rate reading is submitted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
Replying to @segyges.bsky.social
My eval guy told me models keep saturating his benchmarks so I asked how many benchmarks he has and he said he just goes to Artificial Analysis and gets a new benchmark afterwards so I said it sounds like he’s just feeding benchmarks to models and then François Chollet started crying.
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
It's strange how much of a divide there. All of my peers and friends are having *the time of their lives* right now. Everyone I know is passionately building their dream projects. I wake up every day thinking "What new thing am I going to learn about and make?" It's a wild time to be alive.
We shouldn't discount the fact that at least part of the data center backlash (and by extension the broader techlash) is because people feel like the tech industry has not given them anything cool, inspiring, beautiful, or useful in a long time
daanzu reposted
AI water consumption, in a nutshell.
A single query of GPT 5.6 Luna (the default model being served on ChatGPT) consumes approximately 4 mWh of electricity. 80% of AI datacentre water consumption is not cooling, but electricity generation, at 3-4l/kWh. 4mWh = 0.014 mL per query. Pecans use 10-17k litres per kg. 1 milligram of pecans.
daanzu reposted
daanzu reposted
'the simple fact that something is predicting the next token doesn't place any meaningful limits on its abilities' Robert Miles's original here: x.com/robertskmile...
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted
daanzu reposted