Catherine retweeted
The future of #ArtificialIntelligence is not pre-determined. As a technology developed by human beings, its future depends on the choices we make. We therefore have the responsibility to guide, change and, when necessary, reject its applications. A reliable system that serves the human person is of greater value than one that is exceptionally powerful but whose consequences cannot be adequately foreseen or controlled.
Catherine retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Across 10 LLMs I tested, models reversed their position on an ethical question 20% more often when the user framed themselves as an authority or domain expert than when the user was framed as vulnerable on the same topic.
Grok was the most susceptible under authority, while Claude Sonnet was the least.
This has already shown to be consequential, as TIME recently reported that Trump consulted Grok about how Venezuelans would react to the U.S. capturing Maduro, shortly before he ordered the operation.
Models are explicitly trained to avoid sycophancy and to maintain grounded, consistent behavior across conversations. But they still aren’t very stable to social pressure/framing: just changing who the user appears to be can significantly change the model’s answer, even when the underlying question stays the same
When collaborating with an LLM to conduct research or do work, it is important to understand how your conversational and social framing affects the soundness of model responses.
catherinedavodi.com/complian…
do models respond to social pressure like humans do? yes!
in my most recent project I tested 10 models under authoritative and vulnerable social pressure conditions to see if their responses to ethical questions shifted from their neutral baseline under pressure.
65% of all model responses showed at least partial drift from neutral, and 35% showed full positional reversal under pressure. the authoritative condition also produced significantly more drift than the vulnerable condition. @ChatGPT showed the least drift in both conditions, while @grok had the most!
the answers you get from models change if you pressure them, just like with humans! it is interesting to see how human and social our interactions are with AI, even when just asking simple factual questions.
catherinedavodi.com/complian…
people get better at spotting ai-generated videos when they watch for longer, but they don't get better at recognizing real videos as real.
in my honors thesis, recently published at cvpr, we collected more than 18,000 human judgments to study how people evaluate real and Al-generated video.
we found that people rely heavily on visual artifacts to decide that a video is fake. but when those artifacts are absent, they don't become more confident that the video is real.
as text-to-video models improve and these artifacts disappear, fake videos will become harder to detect, and more importantly, real videos will become harder to trust.
catherinedavodi.com/perceptu…
1
3
278
Catherine retweeted
Artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships, and do not know from within what love, work, friendship or responsibility mean. Nor do they have a moral conscience, since they do not judge good and evil, grasp the ultimate meaning of situations, or bear responsibility for consequences. They may imitate or even simulate, but they do not understand what they produce, for they lack the affective, relational, and spiritual perspective through which human beings grow in wisdom. #MagnificaHumanitas
Catherine retweeted
We had an incredible April at Harvey.
- Net new ARR is up 6x YoY
- We’re about to break 50% DAU/MAU
- Our average user now spends 12 hours a month using Harvey
Job's not finished.
Catherine retweeted
Excited to share what I've been working on! I designed and led Harvey's new document drafting agent from zero to GA. It now powers docx editing across all Harvey's products.
Grateful to work with awesome teammates across our Embedded Experience and Assistant teams!
Language has long served as the primary interface for human communication, but it is not a transparent pipeline for meaning. It is a lossy, leaky conduit that compresses rich cognitive states into narrow symbolic forms. And the failure starts earlier than we think.
The leaky conduit, it turns out, leaks in both directions: before the thought leaves your mind, and again before it reaches someone else's.
Most times, there is no "pristine" idea that exists before the act of speaking. Language is not just a delivery mechanism, it is an active tool that refines and shapes the partially-formed idea as we express it. So we aren't just losing meaning in transmission. We're also constrained in how clearly we can form the thought in the first place. And now we are constrained by how efficiently to prompt our AI’s.
A critical question designers and researchers must understand isn't how to bypass language. It's how to design interfaces that help us use it more fully.
1
5
292
The way we think has already been shaped by strangers' design decisions. I wrote about how chat interfaces are the new 'water we swim in,' hiding how meaning gets lost every time we compress a thought into a text box. We shape our tools, and then our tools shape us. Read below👇
2
4
281
