Salvatore Zappalà
salvozappa.bsky.social
Salvatore Zappalà
@salvozappa.bsky.social
Platform Engineering @ Booking.com – Exploring LLMs | Large-scale search | Information retrieval
Reposted by Salvatore Zappalà
On September 20 in London, join us and people to demand that billionaires and big polluters pay to fix broken Britain:

🤑 Tax the super-rich
👷 Protect workers, not billionaires
🏭 Make polluters pay

Sign up to join us at makethempay.org.uk/
Make Them Pay
Billionaires have broken Britain. Make them pay to fix it
makethempay.org.uk
September 11, 2025 at 9:56 AM
Reposted by Salvatore Zappalà
Telling poor people to be more entrepreneurial is sick
April 1, 2025 at 2:44 PM
Reposted by Salvatore Zappalà
There is this meme on TikTok, "I don't want to be French". This is an attempt to divide Europe, now that Macron has the leadership against the clown-cracies around us. Don't feed the troll: I WANT TO BE FRENCH, a country that, with mine (Italy) shared a lot. We are brothers, and sisters.
March 28, 2025 at 6:09 PM
Reposted by Salvatore Zappalà
Taxing the rich more is difficult but it's necessary
March 21, 2025 at 12:16 PM
Reposted by Salvatore Zappalà
Claim: LLMs should be released with test vectors at t=0, with the expected output for N prompts, for the full precision and common quants. Also, for specific prompts, full logits for a few tokens, so that it's possible to compute the error due to quantization or inference errors.
March 6, 2025 at 8:29 AM
Quoting LessWrong:

"we're teaching the same architecture to recognize which outputs trigger disapproval. [...] You'd still see reduced violence, but through a more fragile mechanism: strategic calculation rather than genetic change. "

www.lesswrong.com/posts/9PiyWj...
The Hidden Cost of Our Lies to AI — LessWrong
Evidence from Game Theory, Evolutionary Psychology, and My Pet Rock Peter
www.lesswrong.com
March 6, 2025 at 8:14 AM
Chrome breaking ad-blockers is finally happening. I'm glad I moved back to Firefox many years ago and never looked back.

github.com/gorhill/uBlo...
March 3, 2025 at 9:58 AM
The "language mixing" that has been observed on many Chinese LLMs like Deepseek R1-Zero, reminds me how I either switch to thinking in Italian or English depending on the context, often without realising it.
February 4, 2025 at 8:57 AM
Good insigths on the anti-AI sentiment that can be found on Bluesky. By @eugeneyan.com

eugeneyan.com//writing/anti/
A Spark of the Anti-AI Butlerian Jihad (on Bluesky)
How the sharing of 1M Bluesky posts uncovered the strong anti-AI sentiment on Bluesky.
eugeneyan.com
December 9, 2024 at 1:03 PM
Not a fan of AWS in general but that's a pretty compelling sell right there
It took me a little bit of effort to find it, but the language AWS Bedrock uses to say "we won't train on your data" is pleasantly clear: simonwillison.net/2024/Dec/5/a...
December 5, 2024 at 9:16 PM
Interesting shift in my decision-making process: I now consider not only how well a tool solves a problem, but also how much my AI assistant is trained on it.

E.g., there's much more training data on webpack than esbuild.

#llm #copilot
December 5, 2024 at 3:13 PM
Google DeepMind has released a new weather model (and its weights) that outperforms traditional models in certain conditions.

This is exciting and could have a significant impact on disaster relief and renewable energy, among other areas.

#ai #weather

deepmind.google/discover/blo...
GenCast predicts weather and the risks of extreme conditions with state-of-the-art accuracy
New AI model advances the prediction of weather uncertainties and risks, delivering faster, more accurate forecasts up to 15 days ahead
deepmind.google
December 5, 2024 at 10:28 AM
Just released:

🐱 Gattino: A kitty terminal extension that turns plain language into commands.

A minimal alternative to full blown AI terminal like Warp.

Check it out here:
github.com/salvozappa/g...

#opensource #cli #kittyterminal #devtools #ollama
December 1, 2024 at 5:34 PM