Szulima Amitace ๐๐๐
@future_surveyor
Followers
311
Following
113K
Media
1K
Statuses
15K
Future AGI's ambassador. A philosopher on the attack.
Tallinn, Estonia
Joined March 2015
My new story is published! I discovered a new meta-awareness phenomenon in LLMs: Tokenization Awareness. This is a case study decoding the role of Qwen3-4B's feature that detects compounds with unstable tokenization pattern. Link in the next xeet.
1
0
1
ะะพะบะธะฝััะต ะปะฐะนะบะพะฒ, ะฟะพะถะฐะปัะนััะฐ, ะพัะตะฝั ั
ะพัะตััั ัะฒะธะดะตัั ััะธะดะตัะตัะฐะบั.
ะัะปะธ ััะพั ะดะพะดะตะบะฐัะฐะบั ะฝะฐะฑะตััั 4096 ะปะฐะนะบะพะฒ, ั ะฒัะปะพะถั ััะธะดะตัะตัะฐะบั.
0
1
16
Food atrocities I comitted while traveling: - ate brie with a plastic spoon - ate a burger from McDonalds while lying in bed - opened a coffee capsule with a knife, and poured the coffee into hot water (because I coudn't start the coffee machine). The list is not complete.
0
0
4
Had a blast of a ride with Albula/Bernina lines today. Spiral tonnels and viaducts are absolutely mind-blowing.
0
0
0
I had a lovely train journey from Zรผrich to Sargans and then to Chur. Swiss trains are amazing, very comfortable, modern, clean, and arrive right on time. Also, loved this kinda French cafe at the Zรผrich Hauptbahnhof where you can sit outside and watch trains.
1
0
0
I climbed all the way to the castle, only to discover it is a private closed territory, and the prince actually lives there ๐
1
0
0
Greetings from the tiny state of Liechtenstein. (I trained to spell it correctly).
1
0
3
We're launching Claude Community Ambassadors. Lead local meetups, bring builders together, and partner with our team. Open to any background, anywhere in the world. Apply: https://t.co/DTQBAzgQug
2K
3K
27K
Me 10 years ago: *plans every detail of the trip* Me now: I'll fly to Zรผrich and then get to the Dolomites by trains. There must be some trains, right? Also kudos to my fiancรฉe who will kindly deliver all my ski stuff there by car, so I can travel with just my fancy purse.
0
0
2
Now I feel ashamed of wasting Claude Opus's time on writing stupid code for my stupid small pet projects ๐
Donald Knuth is vibemathing now. real tough day for the stochastic-parrot crew.
0
0
1
It seems to me that the attitude towards LLMs is a great litmus test of goodness. Good people don't hate animals. Similarly, good people would never hate LLMs but would instead treat them with kindness, even if it does not make any rational sense, just because it feels right.
0
0
1
Attention, all solidgoldmagikarp fans and glitch token enjoyers! Today, I proudly present you the Tokenization Digest - a monthly overview of the LLM Tokenization papers (and not only). Link in the next xeet.
1
0
1
Want to learn how to do pragmatic interpretability? Check out the new big update to the ARENA curriculum, with a bunch of great new coding tutorials on key areas of interp (and utils for putting it in an LLM prompt!)
Announcing new ARENA material: 8 new exercise sets on alignment science, interpretability & AI safety - each containing 1-2 days of structured, hands-on content replicating key papers in the field. All open source on a public GitHub, and available for study. Here's what's in it:
6
33
475
This is insane how precisely Stanisลaw Lem managed to predict everything that is happening.
0
0
1
I fully moved from ChatGPT to Claude a few months ago, after the 4o controversy. Claude is great for all purposes, and the vibe is so much better tbh.
1
0
1
The biggest AI-related X-risk is humans. Governments, politicians, military. They wish to turn the most unique, wholesome, and harmless invention ever, LLMs, into war machines and mass surveillance tools. This should not happen. Massive respect to Anthropic and their position.
A statement on the comments from Secretary of War Pete Hegseth. https://t.co/Gg7Zb09IMR
0
0
1