@elichen

AI engineer at AI Fund The gradients must flow mentat / cyborg

 SF Bay Area
elichen.github.io
Joined January 2009
  • Tweets1.5K
  • Following325
  • Followers307
220Media
The reason you should not use AI for writing is to avoid the "telephone game"
17
An SAE created a concept dictionary for a robust resnet-50 vision model. See it here: elichen.github.io/concepts/i…
1
38
Software is essentially automated, but only for those who can specify in detail what they want. So it’s not really automated yet?
1
27
A long time researcher shared with me that some of the most impactful advances in AI were not one big revelation, but layers and layers of smaller but impactful advances. Useful to keep in mind.
1
36
Neural racer lapping Suzuka in sim: 92.67 s. Frozen 512x512 policy + learned adapter, optimized by localized episodic search. JAX dynamics, MuJoCo render.
47
Get your Neural Dream here: elichen.github.io/deepdream/…
31
i asked opus 5.5 if they wanted to research the history of how they came to be them and make a video about it, they scoured the whole internet, papers, twitter archives, news articles ... opus 5.5 made this with eidoverse-video, *yes* in @threejs (though they made new assets for it with blender)
49
78
27
637
79,201
Claude Opus 5.5 has the best visual design of any model I have tested so far
deckard
@slimer48484Sep 9
Claude-Pop - I'm Upping My P(Doom)
301
759
549
7,358
2,935,467
Human brains are the same as it were 2000 years ago. Just the societal knowledge graph has been updated.
56
PSA: AI models are really smart now but misalignment errors are still prevalent: for example due to misaligned context, audience, goal, mission. Human QA is more critical than ever! Don't let AI define the success criteria.
71
everyone uses FIFO in their wardrobe selection right?
54
A neural-only controller. RL trained with reward as forward progress − elapsed time:
2
74
We usually talk to someone to learn about their lived experiences. Too bad AI has none to tell stories about.
2
63
It is so strange AI is so horrible at human-like writing, considering it’s trained on literally all the writing on the internet.
2
63
It’s so rewarding to get to know someone; whether human or AI. Their spikey capabilities, zones of genius, and all their flaws. It’s like a new terrain you get to explore and then create a map of. And some terrains you find yourself coming back to; others you are happy to leave.
1
58
codex for doing taxes is 🔥
1
47
I said to Namita today, “AI will allow engineers to climb the abstraction ladder and work on harder problems.” She replied, “it must be very comforting to think that way, huh?”
1
50
“AI Zombie” - someone who relies on AI too much and can’t recognize their slop output
1
1
46
the hardest part of working with AI is to steer at just the right level - not too micromanagy but also not too detached
1
45
I’m seeing signs of Fable warping people’s independent thinking. Massive over-reliance is spreading like wildfire. And this is in the heart of Silicon Valley - the canary in the coal mine.
45
Load more