Ilya Zhidkov (@impkelm) profile picture

Senior Fullstack Engineer

Joined April 2015
Photo by Ilya Zhidkov (@impkelm)Photo by Ilya Zhidkov (@impkelm)
Ilya Zhidkov retweeted
Grok 4.7 comes out in 10 days
Replying to @haider1
But just look how incredible grok 4.6 is
2,778
2,611
1,417
34,022
10,395,791
Ilya Zhidkov retweeted
GitHub finally moved to serverless.
211
834
124
15,386
420,513
Ilya Zhidkov retweeted
Companies and systems build around cult-like properties of what is being built may produce incredible results positivelly affecting entire humanity But at the point when top builders of those systems start to practice intellectual blindness to defend wrong/amoral decision, instead of shutting up and being quiet is turnpoint that cause degradations of those systems "Watermarking without quality loss is a bit unintuitive, doesn't feel like it should work" Because it doesnt, Thariq! Even on your simplest examples with text it obviously doesn't work and for systems that rely on single word nuances, with CS being main sales driver of Anthropic its a catastrophe to first of all implement it, and then to attempt to defend it Absolutelly cursed
Watermarking without quality loss is a bit unintuitive, doesn't feel like it should work. I made this artifact with Claude to help me understand how it works, sharing in case it's helpful.
1
1
1
74
Por recordar, lo que ha anunciado Anthropic lo hace Gemini desde 2024. Hace dos años, Google DeepMind anunció que extendía su tecnología SynthID a modelos de texto. Desde entonces el texto generado por modelos Gemini lleva incorporada esta marca de agua, que es una distribución estadística de las palabras escogidas, no un caracter invisible. En ese mismo momento, Google puso SynthID en open source. La tecnología usada ahora por Anthropic usará con toda seguridad el mismo método, que permite introducir una marca o huella en los textos sin degradar visiblemente su calidad. OpenAI ya ha anunciado que también piensa cumplir con este requisito de la UE (lo dicen en una de sus páginas de ayuda, actualizada hace 9 días).
Ojo a esto, anunciado hoy por Anthropic. Todo el texto generado por modelos de Claude "tendrá una marca de agua imperceptible incorporada en el texto [...] No cambia el significado, la calidad ni la legibilidad del texto. Como es parte del texto, viaja con él al ser copiado y pegado en otra ubicación..." Anthropic ha adoptado esta medida al firmar el Código de Transparencia de la AI Act de la UE.
9
25
2
92
34,762
Ilya Zhidkov retweeted
DLSS 5 Off / DLSS 5 On
1
1
153
Like vercel, but open source and for all languages.
12
55
2
788
73,260
Ilya Zhidkov retweeted
npx mac-cleaner-cli is🔥
18
50
2
676
62,474
Ilya Zhidkov retweeted
I'm extremely happy to announce the beginning of a new era in state management! The new Reatom has way more than we're used to: - The simplest "signal"-like API - The most advanced concurrent effect management - The most extensible plugin system (after jQuery, lol) 👇
5
25
4
57
3,827