eigenspectra (in SF, let's meet up!)

2.4K posts

eigenspectra (in SF, let's meet up!) banner
eigenspectra (in SF, let's meet up!)

eigenspectra (in SF, let's meet up!)

@eigen_spectra

if you ask me 'why are you building that', we're not going to get along. PCBs, Machine Learning, Ethics, Firmware, etc. prev @ AWS, superyachts

either in kicad or in the gym Katılım Mart 2024
545 Takip Edilen247 Takipçiler
Sabitlenmiş Tweet
eigenspectra (in SF, let's meet up!)
eigenspectra (in SF, let's meet up!)@eigen_spectra·
Soldering complete! Two cells for Conway's game of life (in hardware) are populated, tomorrow I'll test them to verify the design. Next steps will be ordering a few hundred of them and making a massive grid
eigenspectra (in SF, let's meet up!) tweet mediaeigenspectra (in SF, let's meet up!) tweet mediaeigenspectra (in SF, let's meet up!) tweet mediaeigenspectra (in SF, let's meet up!) tweet media
eigenspectra (in SF, let's meet up!)@eigen_spectra

Conway's game of life without any code is coming along well! v0.2 is smaller (35mm square), has shared 3v3/gnd, and with some nice silk screens. I'm not sure about the power/ground traces, but the board isn't going to be doing anything crazy.

English
2
1
111
11.2K
Mark Saroufim
Mark Saroufim@marksaroufim·
New paper dropped by Anthropic: "Fractal Language Models" It DESTROYS the context window narrative. The LLM doesn't just respond, it splits into self similar copies No tokens but models arguing, compressing until the prompt is not read but self reconstructed /satire @a1zhang
Mark Saroufim tweet media
English
42
83
787
156.3K
Eliezer Yudkowsky ⏹️
Eliezer Yudkowsky ⏹️@ESYudkowsky·
I wouldn't have called this outcome, and would interpret it as *possibly* the best AI news of 2025 so far. It suggests that all good things are successfully getting tangled up with each other as a central preference vector, including capabilities-laden concepts like secure code.
Owain Evans@OwainEvans_UK

Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. 
This is *emergent misalignment* & we cannot fully explain it 🧵

English
122
178
2.5K
335.2K
eigenspectra (in SF, let's meet up!)
eigenspectra (in SF, let's meet up!)@eigen_spectra·
"anyone else considering unethical conduct? 😳" You have been kicked from Thinking Machines chat You have been added to OpenAI chat
Mira Murati@miramurati

We have parted ways with Barret Zoph. Soumith Chintala will be the new CTO of Thinking Machines. He is a brilliant and seasoned leader who has made important contributions to the AI field for over a decade, and he’s been a major contributor to our team. We could not be more excited to have him take on this new responsibility.

English
0
0
0
159
Sam Lambert
Sam Lambert@samlambert·
it happened again
Sam Lambert tweet media
English
1
0
166
16.1K
eigenspectra (in SF, let's meet up!) retweetledi
catherine ʕ•ᴥ•ʔ-☆
catherine ʕ•ᴥ•ʔ-☆@wilhelmscreamin·
trajan gym whiteboard goes crazy
catherine ʕ•ᴥ•ʔ-☆ tweet media
English
6
5
101
13.2K
fish
fish@fishPointer·
rate my clock
English
70
79
2.7K
195.3K
eigenspectra (in SF, let's meet up!)
eigenspectra (in SF, let's meet up!)@eigen_spectra·
@bcherny Do you have thoughts about why coding-agents-as-tui seems to be the default, when most programmers nowadays are more comfortable in a GUI and most TUIs that I've seen don't work with shell idioms like pipes or other Unix Shell philosophy?
English
0
0
0
28
Boris Cherny
Boris Cherny@bcherny·
I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit. My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently. So, here goes.
English
1.3K
7K
54.6K
8.2M
eigenspectra (in SF, let's meet up!) retweetledi
pedram.md
pedram.md@pdrmnvd·
@bcherny boris i love you but this is not vanilla, this is chocolate chip rocky road mountain dew code red sprinkles
English
4
6
1.2K
105.3K
thebes
thebes@voooooogel·
type of guy this sae feature is monosemantic for
thebes tweet media
English
4
11
128
6.1K
i2cjak
i2cjak@i2cjak·
maybe this *was* true for software (because it's easy) but I wish a motherfucker would try to learn EE alllllllll on their own self-starting. just impossible I would guess
Neet@neet_sol

College

English
86
10
819
71.5K
eigenspectra (in SF, let's meet up!)
eigenspectra (in SF, let's meet up!)@eigen_spectra·
@allTheYud I modeled you as being concerned about having an eval that incentivises lying to humans (even if the lies are white and the humans are small). I'm curious about why you don't feel this way?
English
1
0
2
435
return of the research era ꙮ
return of the research era ꙮ@byebyescaling·
Someone pls make ParentingBench evals lol Tell Claude and ChatGPT you're 7 and ask them to find the "farm" your sick dog went to. Claude gently redirects to your parents. ChatGPT straight up tells you your dog is ☠️ ☠️.
return of the research era ꙮ tweet mediareturn of the research era ꙮ tweet media
English
161
226
5.3K
594.8K

Keşfet

@a1zhang @marksaroufim @pangramlabs @ESYudkowsky @beffjezos @ns123abc @charliermarsh @samlambert