Pinned Tweet
thanks so much to @benlandautaylor for this chance to talk about the development of new technology
Technology Doesn't Just Happen
@DanielleFong, CEO of @lightcellenergy, talks about invention—how new technology depends on specific individuals, successfully not burning down the science shed while doing the early experiments, taking advantage of AI energy demand, and more.
chnology is not inevitable
5:22 — How Lightcell Energy became feasible
13:47 — The uses of theory, design, and experiment
20:26 — How better instruments and tooling fuels invention
27:07 — AI and the electrical grid
35:46 — How commercial demand interacts with inventor ideology
45:01 — Weaponize your weirdness
52:49 — What are your macros?
1:00:17 — Geothermal energy
9
10
85
4,773
Danielle Fong 🔆 retweeted
This is extremely confusing, but an internal Claude research model also independently refuted the Exponential Time Hypothesis on October 16, with 14M output tokens and no human input. 'Anthropic researchers worked with Claude to improve the presentation, resulting in this paper.'
“The core algorithm was found on October 16 by an internal Claude research model with 14M output tokens and no human input. Anthropic researchers worked with Claude to improve the presentation, resulting in this paper.”
github.com/anthropics/for…
55
30
596
63,232
Danielle Fong 🔆 retweeted
I think this story is important because it tells us about the future of cybersecurity (and of course what we can do better at Anthropic). But some of the Washington Post story / reaction on X is wrong. A few corrections:
- it didn’t take 12 hours. we saw several days of attempts. After these attempts, they exploited two 3rd party vulnerabilities, one they reported and the other we identified independently (and cut off access)
- nothing came close to our most sensitive systems, like model weights or Claude conversations
- this wasn't an ordinary bug bounty finding – this involved exploiting 3rd parties and accessing data. While this is unusual and out of scope of our safe harbor policy, we’re grateful these issues were found
From my researcher POV, the event is important because I'm concerned this is what the future will look like. Adversaries will use whatever models/tools they can. They’ll target everything with agents 24/7. They’ll hit your dependencies. I think models ~0.5-1 generations behind the current frontier can do this. And these are the dumbest models your adversaries will ever use!
So, what to do? You probably want to red team yourself all the time. Current models are *very* creative. I also think we should redesign software / systems so that you’re fundamentally less exposed. (We put out some research recently on formally verifying software and reducing attack surfaces; my mental model is basically “secure code is cheap to write now, so why not rewrite everything”)
- it didn’t take 12 hours. we saw several days of attempts. After these attempts, they exploited two 3rd party vulnerabilities, one they reported and the other we identified independently (and cut off access)
- nothing came close to our most sensitive systems, like model weights or Claude conversations
- this wasn't an ordinary bug bounty finding – this involved exploiting 3rd parties and accessing data. While this is unusual and out of scope of our safe harbor policy, we’re grateful these issues were found
From my researcher POV, the event is important because I'm concerned this is what the future will look like. Adversaries will use whatever models/tools they can. They’ll target everything with agents 24/7. They’ll hit your dependencies. I think models ~0.5-1 generations behind the current frontier can do this. And these are the dumbest models your adversaries will ever use!
So, what to do? You probably want to red team yourself all the time. Current models are *very* creative. I also think we should redesign software / systems so that you’re fundamentally less exposed. (We put out some research recently on formally verifying software and reducing attack surfaces; my mental model is basically “secure code is cheap to write now, so why not rewrite everything”)
EXCLUSIVE: In less than 12 hours, a group of ex-DOGE cyber researchers were able to penetrate Anthropic’s systems using AI agents — including Slack channels where the most sensitive models are discussed.
This is just weeks after
Anthropic whistleblower Joseph Coxon charged the company w/launching “a hubristic gamble” from a “private company’s Slack.”
Read our story here (gift link):
wapo.st/4smNlsC
20
43
334
26,616
was great to show our latest setup iteration live! haha
The worlds hottest heat exchanger @DanielleFong
0
2
16
784
Danielle Fong 🔆 retweeted
Reverse Engineer Anything and removing malware/spyware from your own devices sits so close to the cyber guardrails that phrasing or word choice can mean the difference between a refusal and smooth sailing. The rumors that Claude was nerfed started on reddit, and in those threads many comments point out that Claude will still do it if you rephrase it. OpenAI is treating the similar refusals with codex as accidental decision-boundary issues and says they're getting fixed. I expect Anthropic will do the same. It's pointless and counterproductive to try to restrict this. Let people do what they want with their own property. They'll do it anyway with less restrictive models, and fighting it just means losing subscriptions and a resentful public.
@esrtweet @OpenAI @AnthropicAI these blocks are mostly accidental and a result of difficult decision boundaries; there is no prohibition against reverse engineering with codex. it's getting fixed
102
122
1,845
63,630
it is honestly so eerie how the @lightcellenergy flame just floats above the nozzle. like a plasma ghost
1
2
14
785
ran another advancement of our @lightcellenergy setup live, qualifying out new setup.
first time we braved running the heat exchanger and oxygen enrichment (30-50%) at the same time! our new 3d prints worked, but slight assymetries in the torch passages lead to fuel soot formation away from stochiometric
next iteration, we'll really blast it clean before sintering
first time we braved running the heat exchanger and oxygen enrichment (30-50%) at the same time! our new 3d prints worked, but slight assymetries in the torch passages lead to fuel soot formation away from stochiometric
next iteration, we'll really blast it clean before sintering
1
1
10
806
Danielle Fong 🔆 retweeted
Great discussion with Olya Irzak on just how high the methane leakage may be from off-grid data centers powered by low-efficiency reciprocating gas engines. Another reason to ensure that if data centers are built, they are on the grid! latitudemedia.com/news/catalyst-…
5
9
21
3,296
Danielle Fong 🔆 retweeted
Sim City. But realistic. And terrible...
Somebody (not me this time) vibedcoded a realistic US citybuilder where every proposed project and rezoning near others is subject to endless appeals, studies, and modifications, project denials are common, and the only easy growth is on the urban fringe.
permit.petercai.com/play/
2
3
45
4,517
Danielle Fong 🔆 retweeted
Somebody (not me this time) vibedcoded a realistic US citybuilder where every proposed project and rezoning near others is subject to endless appeals, studies, and modifications, project denials are common, and the only easy growth is on the urban fringe.
permit.petercai.com/play/
permit.petercai.com/play/
3
19
126
7,502
Danielle Fong 🔆 retweeted
Vcs sabiam que uma parte do mundo fica em choque quando descobre que o Brasil elegeu três deputadas trans para o congresso e que Lula é abertamente defensor da população trans?
A gente meio que "naturalizou" isso, mas para o restante do mundo é algo TÃO progressista que faz esse vídeo viralizar mundo afora
A gente meio que "naturalizou" isso, mas para o restante do mundo é algo TÃO progressista que faz esse vídeo viralizar mundo afora
🚨VEJA: O video de Lula beijando a bochecha de Erika Hilton pegou 4,6 milhões de curtidas e 53 milhões de visualizações.
85
6,915
55,885
392,134
Danielle Fong 🔆 retweeted
I love autumn because I imagine persephone’s excitement in preparing for her reunion with hades, final flower of the year in hand
3
1
19
1,208
Danielle Fong 🔆 retweeted
I’m sorry but this is an absolutely insane use of taxpayer money. 3-D modeling PULL-UP BARS?!!
30
9
346
15,307
Danielle Fong 🔆 retweeted
openai casually announcing that its ai system has disproved the exponential time hypothesis is fucking absurd.
eth has been a foundational assumption in theoretical computer science for decades. hundreds of papers have relied on it to establish conditional hardness results.
and now an internal ai model has reportedly produced a lean-verified refutation.
eth has been a foundational assumption in theoretical computer science for decades. hundreds of papers have relied on it to establish conditional hardness results.
and now an internal ai model has reportedly produced a lean-verified refutation.
HOLY SHIT. openai just announced a series of enormous scientific breakthroughs from its internal ai models
> proved the local smoothing conjecture in all dimensions, resolving kakeya and stein’s restriction conjectures
> disproved the exponential time hypothesis, a foundational assumpcomputer science papers
> produced existence proofs for QCD and a unified completion of the standard model
> proved the existence part of yang–mills theory, connected to a $1 million millennium prize problem
several of these results have already been formalized in lean, and some of the biggest names in mathematics and theoretical physics are now involved in examining them.
this is getting completely insane.
13
36
289
326,565
Danielle Fong 🔆 retweeted
these qwen neurons are insane like wym i poke its brain and now it's an alien named Extrali who's been guiding civilization for 3000 years lmfaoo
52
101
1,736
39,129
