Guglielmo Camporese

1/7 recently i started exploring mechinterp, and i started with audio diffusion. i spent some time cracking open stable audio 3 to see if mood has a zip code in its residual stream. it does! it's layer 11 and you can dial it. (audio results below) 🧵

shoutout to @StabilityAI @stableaudio for open-sourcing stable audio 3 - clean model, clean representations, very fun to break open 🙏

7/7 more on my blog and code! - blog: guglielmocamporese.github.io/blog/audio-moo… - code: github.com/guglielmocampo…

1/7 recently i started exploring mechinterp, and i started with audio diffusion. i spent some time cracking open stable audio 3 to see if mood has a zip code in its residual stream. it does! it's layer 11 and you can dial it. (audio results below) 🧵

@antirez @ggerganov goat!

Breaking! @ggerganov himself confirming that DS4 2 bit quants actually work very well! He is running AIME2025, but check his words here in the screenshot.

@k0k1eth you're missing restaurants/apero, sport, house/apartment insurance, and public transport pass

Approximately average minimum living costs in Switzerland: - 2000 CHF rent - 1000 CHF healthcare - 100 CHF internet + mobile - 600 CHF groceries - 50 CHF electricity Total: 3650 CHF On top of this comes car leasing etc if you have one. And you will earn more working at McDonalds than taking the average crypto payment.

@GianlucaZanell2 chatgpt

Ma i vari opinionisti e criminologhe che venerdì sostenevano di aver già letto tutta l'informativa che super potere hanno? Sarà che non sono studiato ma sono ancora a metà #Garlasco

@abakcus 2d fourier juggler trasform

Juggler juggling juggling jugglers.

@DavidDuvenaud @AlecRad @status_effects love how neat model.py is, just 300 lines, plain torch code!

Announcing Talkie: a new, open-weight historical LLM! We trained and finetuned a 13B model on a newly-curated dataset of only pre-1930 data. Try it below! with @AlecRad and @status_effects 🧵

@nic_amadio after two years in switzerland (zurich), I still haven’t recovered from how much I spend on eating out and groceries lol :/ - and I lived mostly in italy (padova), and in the bay area (palo alto).

A few considerations after starting to do online grocery in Switzerland: 1. Salads in CH start from 1.5/2 CHF per 100 grams WTF 2. Coop delivery starts from 100 CHF, no delivery fee after 200 CHF For 2 people: 100 CHF is unfortunately too easy to reach, so not an issue ordering online really. Over 200 CHF: requires a bit of planning. So far I've done 2 grocery orders both around 150 CHF. 3. Beef, we already talked about it (good) 4. Deals You get significantly fewer deals than by going in person at the supermarket. Which is a bit of a shame 'cause one good aspect of Swiss supermarkets is that you can find good offers, since the mass/industrial production (where discounts happen) manages to also retail good quality/organic (although not always tasty) stuff. 5. Makes you realise how 50-80% of the stuff is completely overpriced compared to South/Eastern Europe Is it better than South/Eastern Europe? Not really. - Organic/bio certifications are solid in CH, but also good in EU - Over half the stuff is imported from Southern Europe anyways - When you import, you usually pick fruits/vegs before they're ripe (so they can stay in shape during/after transport), which is also why produce is often watery/tasteless in CH - Some imports are not good in CH, 'cause it's harder to import some stuff in CH than in EU: i.e. SALMON IN CH SUCKS, as does most seafood and fish --- In general: pretty mid/bad experience. One good remark I can mention is that, at least, in rural CH (which could be an interesting option as mentioned yesterday), it can only get better: - more access to local produce (hopefully better prices too) - worst case: Coop/similar still ship to any small town in CH After dropping 300 CHF in Coop groceries in just 2 days, gf couldn't hold a: "let's move to Croatia" 😄 (I pay grocery in CH, and we can afford it, but it still feels like burning money for no reason) Heard from a friend who had family with few kids in CH that they spend 2-3k CHF/mo mostly for just grocery. Many pros to being in CH even as on online/remote worker. But damn sometimes it's hard to bite: - restaurants/bars/cafes/activities all expensive af - even grocery expensive af for mid stuff in 80%+ of categories Good public schools for kids but also 10-15k+ CHF to put them braces/Invisalign. Good healthcare but also 300-700 CHF/mo per person depending on where you live (a bit less for kids). Geopolitically safe but also how likely are you to die in Croatia because of some war?

@kalomaze is it like a rvq-vae trained with discrete tokens having hilbert structure?

hilbert curve waveforms

@francoisfleuret I see reasoning as iteratively focusing on what matters, discarding what doesn’t, and narrowing the solution space :)

What is reasoning? Stuff the model can generate and is such that conditioning on it allows the model to generate good stuff it can otherwise not generate?

@LiorOnAI “It’s over” blablabla

It's over. Karpathy just open-sourced an autonomous AI researcher that runs 100 experiments while you sleep. You don't write the training code anymore. You write a prompt that tells an AI agent how to think about research. The agent edits the code, trains a small language model for exactly five minutes, checks the score, keeps or discards the result, and loops. All night. No human in the loop. That fixed five-minute clock is the quiet genius. No matter what the agent changes, the network size, the learning rate, the entire architecture, every run gets compared on equal footing. This turns open-ended research into a game with a clear score: - 12 experiments per hour, ~100 overnight - Validation loss measures how well the model predicts unseen text - Lower score wins, everything else is fair game The agent touches one Python file containing the full training recipe. You never open it. Instead, you program a markdown file that shapes the agent's research strategy. Your job becomes programming the programmer, and this unlocks a strange new loop: 1. Agents run real experiments without supervision 2. Prompt quality becomes the bottleneck, not researcher hours 3. Results auto-optimize for your specific hardware 4. Anyone with one GPU can run a research lab overnight The best AI labs won't just have the most compute. They'll have the best instructions for agents who never sleep, never forget a failed experiment, and never stop iterating.

@giffmana @natolambert Man, Zurich is sooo boring though! There’s no local music scene, the food is meh, restaurants close super early - but yeah, the services are great (you get what you pay for)

@natolambert Yes but the lake is very different, plus we have a well functioning society. I regularly say, Vancouver is the one place i will happily move to, if Switzerland ever implodes. Seattle not so much though.

What a beautiful late winter day. Too bad my phone camera can't do it justice:

@giffmana 🇮🇹🇨🇭

They are really putting AI into everything these days...

@sainingxie and maybe less cold?

<AI global warming>: there may be several AI winters, but they keep getting shorter.