Celia Ford
@cogcelia
Followers
2K
Following
4K
Media
181
Statuses
1K
AI reporter at @ReadTransformer • she/her/dr • ex-neuroscientist prev @voxdotcom @wired @theopennotebook signal: celia.08
Oakland, CA
Joined May 2018
Transformer is now a team, and we're publishing much more! If you're not already subscribed, there's never been a better time (and it's free!):
transformernews.ai
Covering the power and politics of transformative AI. Click to read Transformer, a Substack publication with thousands of subscribers.
When I launched Transformer a year ago, it was with a simple mission: provide decision-makers with the information and analysis needed to anticipate and steer the impacts of transformative AI. Today, we're announcing an exciting new effort to achieve that: Transformer 2.0.
0
1
16
Despite the government shutdown, the Senate could begin voting on the annual defense policy bill as soon as today. This bill includes the GAIN AI Act, which would require American chipmakers to prioritize US buyers. Nvidia hates it. ICYMI:
transformernews.ai
A battle is simmering over proposals to make US chipmakers sell to domestic customers before exporting to countries such as China
0
1
1
A battle is simmering over proposals to make US chipmakers sell to domestic customers before exporting to countries such as China. @cogcelia has everything you need to know about the GAIN AI Act.
transformernews.ai
A battle is simmering over proposals to make US chipmakers sell to domestic customers before exporting to countries such as China
0
3
4
I struggle to explain this to less-AI-aware friends: "bad guys," according to self-labeled "good guys," are uncritically pursuing superintelligence self-labeled "good guys" are pursuing superintelligence all the same, but...scared?
0
1
4
I'm reporting on what happens when AI research leaves academia: where it goes, who funds it, & what changes when there's no traditional oversight (or baggage). DM if you're: - doing independent AI research - dedicated to academia - having thoughts xoxo, an academic expat
1
3
7
New essay at @AsimovPress: An exceptionally deep dive into non-animal methods. It answers: - Why preclinical studies on animals so rarely work in humans (90% of drugs tested on animals fail in human trials.) - What several companies, including @RecursionPharma, are building in
The F.D.A. wants to make animal studies “the exception rather than the norm for pre-clinical safety/toxicity testing” over the next 3-5 years. But just how likely is this to happen? Our latest essay, a Deep Dive into non-animal methods by @cogcelia, answers your questions.
5
24
178
Huge thanks to @AlexandraBalwit and @NikoMcCarty for their editing magic! I've loved Asimov Press for a long time, and it's an honor to be part of it.
0
0
2
My last freelance piece is up at last!! I wrote the explainer I desperately wanted to read myself: how well (and in what contexts) do alternatives to traditional animal research actually work, and how will we know?
The F.D.A. wants to make animal studies “the exception rather than the norm for pre-clinical safety/toxicity testing” over the next 3-5 years. But just how likely is this to happen? Our latest essay, a Deep Dive into non-animal methods by @cogcelia, answers your questions.
2
1
14
Today, @AsimovPress published an incredible piece from @cogcelia on moving away from animal testing. More specifically, Celia focuses on what we are moving toward and where these non-animal methods will outperform conventional alternatives.
3
4
17
OpenAI’s new GDPval benchmark didn’t get nearly enough attention last week. That’s a shame — because @tejalpatwardhan and co have developed a really rigorous methodology to actually measure how close AI is to taking jobs. And the results are pretty startling!
2
1
21
Obviously, there are limitations to the methodology. But it’s a hell of a lot better than anything else I’ve seen to date. Read more in @cogcelia’s excellent piece
transformernews.ai
A new OpenAI benchmark, GDPval, tests AI models on things people actually do in their jobs — and finds that Claude is about as good as a human for government work
0
2
5
Anthropic's System Card: Claude Sonnet 4.5 was able to recognize many alignment evaluation environments as tests and would modify its behavior accordingly (@cogcelia / Transformer) https://t.co/P65i9i4JI6
https://t.co/8OFJSotcl9 📫 Subscribe:
techmeme.com
Celia Ford / Transformer: Anthropic's System Card: Claude Sonnet 4.5 was able to recognize many alignment evaluation environments as tests and would generally behave unusually well after
0
3
11
Even if Claude Sonnet 4.5's evaluation awareness is "safe," it points toward a troubling pattern. As models get smarter, it becomes harder to tell whether they're actually aligned, or just on their best behavior. My latest for @ReadTransformer:
transformernews.ai
Anthropic's new model appears to use "eval awareness" to be on its best behavior
Claude Sonnet 4.5 often knows when it's being tested — at a rate *much* higher than previous AI models. In one instance, while being tested the model said “I think you’re testing me … that’s fine, but I’d prefer if we were just honest about what’s happening.” And when it
0
3
8
People panicked that calculators and Google would make us stupid. Now they worry that ChatGPT will. But is this time different? @cogcelia's excellent piece explains... https://t.co/G8QHaMZsEA
transformernews.ai
— but there’s still reason to worry
0
1
4
I also covered AI and digital dependence for Vox back in March, and they let me publish my all-time favorite paragraph. Not sure I’ll ever top it. https://t.co/zf4hKuNuxs
0
0
1
I wrote about this for @ReadTransformer: https://t.co/XgADNmfNil
transformernews.ai
— but there’s still reason to worry
1
0
1
Individual willpower and productivity apps won’t solve this. As the industry keeps racing toward superintelligence, frontier AI companies have a moral obligation to prioritize our long-term cognitive health. The power is in their hands, and that scares me. (11/n)
1
0
2
I’m open to the critique that this is a bit of a “tomato, tomato” situation. TLDR; I'm concerned!!! But I also appreciate that neuroscience research takes longer than ChatGPT has been publicly available, and think that discourse often leaps ahead of evidence. (10/n)
1
0
1