charles (@filailing) header
charles (@filailing) profile picture

SWE | AI research & ethics | Decentralized governance & economy wanter

Austin, TX
Joined January 2013
Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)Photo by charles (@filailing)
Manual audits assume a reviewer can catch what an attacker would find. That assumption doesn't survive an attacker with more compute than your defense. Proving code mathematically matches intent does. NEAR Co-Founder @ilblackdragon on why formal verification is needed.
The security model the whole industry runs on just broke. For decades it was smart people audit the code and hope the attackers don't find what they missed. Now the attacker is an AI with more compute than your defense, and the only real answer is to mathematically prove the code does what it says. @ilblackdragon on the paradigm shift, and why @NEARProtocol is building its roadmap around proof.
13
40
2
338
129,365
IMHO, i feel the proper audience for Polkadot is companies, not the retail. Polkadot is about infra. We want users that don’t even know they are are on Polkadot (Mewe, @origin_trail, @playmythical …). Sports sponsorship, influencers, taxis = just a BIG waste of money…
2
2
9
189
Very excited to have been elected as Polkadot Head Ambassador! There is lots to do: - set up the new Ambassador program with the other HAs - build a culture of growing strong Polkadot representatives that cover all regions and verticals - run campaigns that have punch
I am running for Polkadot Head Ambassador My main goals are 1. to build up a great culture among the ambassadors to help each other represent Polkadot 2. ramp up video production for agenda-setting and narrative development opengovwatch.notion.site/Pol… Most of my DOT is staked. Is someone willing to put up the 5k DOT decision deposit? I can compensate for the opportunity cost.
59
15
8
293
13,996
charles retweeted
A rare picture of @gavofyork giving @VitalikButerin signed JAM gray paper. What a time to be alive. What can we expect next? 🧐
11
57
13
301
21,912
My first PhD paper!🎉We learn *diffusion* models for code generation that learn to directly *edit* syntax trees of programs. The result is a system that can incrementally write code, see the execution output, and debug it. 🧵1/n
111
577
125
5,308
742,942
This past week was an incredible time at @ETHBerlin! We had our very own Launch Event at the Moos Space (with @VitalikButerin stopping by) and we created a site to track donations to ETH Evacuations during a live interview. Here’s a summary of what happened this past week. 🧵
2
2
1
30
3,126
charles retweeted
pov you took a 1000 mg edible before going to your job as a trader joe’s cashier
170
3,845
251
43,432
2,995,703
Former OpenAI board member Helen Toner: left to their own devices, AI companies might end up like social media, fighting for user attention and consolidating power
50
28
26
295
328,796
I joke, but for a long time this has been the predictable outcome of the constant arms race between spammers and service providers. I support this approach as the only long term scalable solution, and am glad someone is finally willing to bite the bullet. More will follow.
Another nail in the coffin.
21
6
4
146
10,760
charles retweeted
We must accelerate a solution to political stupidity.
68
41
5
479
57,590
charles retweeted
one day PhDs will animate every object around us with reinforcement learning to keep their thesis going
56
330
93
3,244
464,226
This can be somewhat subtle, e.g. "AI could erode fundamental rights", "AI could lead to concentrations of power", etc. Rule-of-thumb: whenever you say "AI could" (or similar), ask yourself if you're confident it's not already happening.
3
2
23
1,061
.@gavofyork when asked, "What's your handy elevator definition of Web3?"
22
46
5
259
28,766
charles retweeted
only the cool kids understand
53
2,038
110
13,391
288,135
Today we’re introducing SceneScript, a novel method for reconstructing environments and representing the layout of physical spaces from @RealityLabs Research. Details ➡️ bit.ly/3x2cOzh SceneScript is able to directly infer a room’s geometry using end-to-end machine learning and represent it using language. Compared to previous approaches, this results in representations of physical scenes that are compact, complete, interpretable and extensible.
44
483
100
2,588
334,608
A look at how Al can alter a movie dialogue and even languages
545
3,024
706
29,550
8,126,190
charles retweeted
how do people still not understand about AGI?? they will say oh it's not actually intelligent. or it's technologically impossible. or it's not even a real concept philosophically. they refuse to think about the curve of progress. skate where the puck is going folks goddamn
24
11
2
154
15,282
It's inevitable that gen AI companies will try to get these court cases dismissed. But I still find it sickening when they do. Some highlights from OpenAI's motion to dismiss the NYT's case: - They claim the NYT 'hacked' ChatGPT. This is PR. There was no 'hack'. The NYT used ChatGPT exhaustively to show that it regurgitates content. - They say ChatGPT doesn't compete with the NYT. This will be hard to prove one way or the other on short timelines (gen AI is still in its infancy), but it's clearly very likely that gen AI will compete with the data it's trained on. - They present a view of gen AI training being fair use that is one-sided. This isn't surprising. In reality the question is much more nuanced than they suggest, and it's yet to be decided by the courts. - They say AI companies have invested billions of dollars based on 'clear and longstanding' legal principles that suggest gen AI training is fair use. I would be very surprised if this were true of any gen AI company out there. Gen AI companies who train on copyrighted work without consent are generally fully aware of the risks. Besides, you can't say 'we invested lots of money in this because we thought it was fair use, and that proves it's fair use'. Other interesting things to note: - They argue that some counts should be dismissed because some of the data scraping + training happened 3+ years ago. This is yet another reason it's important courts + legislators move fast. - In part of their defense, they cite the fact that NYT journalists have praised ChatGPT. This should be a warning to journalists everywhere.
6
94
3
294
13,879
📢 Excited to deliver an oral presentation at #NeurIPS2023 Workshop on CRL tomorrow! 📄 The Linear Representation Hypothesis and the Geometry of Large Language Models (arxiv.org/abs/2311.03658) 🕒 09:50 am - 10:05 am (Poster: 10:30 am - 12:00 pm) 📍 Room 243 - 245
1
10
1
47
9,743
Actually, the Gemini debacle showed how AI ethics *wasn't* being applied with the nuanced expertise necessary. It demonstrates the need for people who are great at creating roadmaps given foreseeable use. I wasn't there to help, nor were many of the ethics-minded ppl I know.
For those who seem to be confused: “AI safety” (as a field) has nothing to do with the woke Gemini debacle. That is a result of “AI ethics” - a completely different thing: AI ethics: focussed on stuff like algorithmic bias. Very woke & left-leaning. Dislike transhumanism & EA & e/acc. Have historically been dismissive of AI safety ppl for “distracting” from their pet ethics issues. AI safety: typically focussed on deep mathematical/game theoretic issues like misalignment & catastrophic risks from future AI systems. Often transhumanist/long-term focussed. Not woke. Spans across political spectrum. For some reason the two groups are getting conflated - in part because certain bad faith “accelerationists” have been strategically using “AI safety” to describe both groups (because they hate both) — but be aware they are VERY VERY different, both in terms of politics, the problems they care about, and general vibe. Don’t get hoodwinked by those who deliberately try to conflate them.
33
58
12
326
97,122