Seth Stafford
@seth_stafford
Followers
805
Following
13K
Media
423
Statuses
3K
Mindsmith forging simple minds for practical tasks at https://t.co/HbBu33ynny. DL is condensed matter physics in theory, but metallurgy in practice. @sethts.bsky.social
San Carlos, CA
Joined June 2009
Our joint announcement with athenahealth Marketplace https://t.co/2S0tyBNHxQ
0
1
2
ThetaRho, a leader in AI-powered intelligence systems, is proud to announce our partnership with athenahealth, Inc by releasing our newest AI application, RISA, to the athenahealth Marketplace. #Agentic #AI #data #Healthcare #FHIR
0
1
2
Daniel Dennett's warning against Counterfeit People made some of the same points. https://t.co/cnpClDTHgb
theatlantic.com
Companies using AI to generate fake people are committing an immoral act of vandalism, and should be held liable.
What I call Seemingly Conscious AI has been keeping me up at night - so let's talk about it. What it is, why I'm worried, why it matters, and why thinking about this can lead to a better vision for AI. One thing is clear: doing nothing isn't an option. 1/
0
0
1
I'm breaking my silence. For years, I was quiet about how @EricRWeinstein, Sabine Hossenfelder, @DrBrianKeating & @TOEwithCurt suppressed my scientific critique. They preach free inquiry but practice censorship. This is the story of their hypocrisy. 🧵 https://t.co/6h6TZsIlB7
156
301
3K
Bingo ... the "last mile" problem is the key issue in building AI-powered apps. The crowd all believe AI is only a matter of scale, but people building apps know there are challenging long tail (or "last mile") issues. Domain semantics matter. IYKYK @ThetaRho_ai
My hot take: The vertical AI products that will win are not the ones with the most sophisticated models or techniques - but those with the best system for incorporating domain expertise. The LLMs in the product need to understand the industry-specific and customer-specific way
0
1
1
"Potemkin Understanding" ... when a model 'understands' a question well enough to provide a correct answer, but a follow up question quickly reveals that the model can't *apply* that 'understanding' All too human!
“Potemkin Understanding in Large Language Models” A detailed analysis of the incoherent application of concepts by LLMs, showing how benchmarks that reliably establish domain competence in humans can be passed by LLMs lacking similar competence. H/T @acowley@mastodon.social
0
0
1
Brisk, accessible, yet wide-ranging talk. 1-slide 'proofs' of Gödel's incompleteness and Turing halting problem results. (They are simpler ideas than they are often made out to be.)
another great lecture by Scott Aaronson, on, loosely: - the nature and limits of computation, proofs - the busy beaver function and trying to quantify uncomputability - some speculations on hypothetical computing models vs the laws of physics - many other interesting tidbits
0
0
1
Great experiment and visualization. Watch gradient descent learn a discrete algorithm!
To understand what is happening in the RNN, we extract automata from its hidden representations during training, which visualize the computational algorithm as it is being developed. (5/11)
0
0
3
Everyone's chasing AI talent from the obvious places. But here's the thing: you can't build another OpenAI by just building an LLM. If you want to replicate OpenAI's success, maybe don't just go after the latest employees from top labs. Go a step further and look at who trained
5
17
99
Being a scientist is still the best job in the world. The constant frustrations, all the crap that goes with trying to keep the lab alive and move things forward - it all resets and washes away in those occasional moments when we catch a glimpse of nature revealing something
77
130
1K
Fun discussion, especially the physics/AI analogies. Highly recommended!
thanks for having me on steve. good discussion !
0
0
1
Agentic Interpretability ... interesting new direction. Artificial psychoanalysis! But are humans the analyst or the analysand? Maybe a bit of both.
We (@_beenkim @johnhewtt @NeelNanda5 Noah Fiedel Oyvind Tafjord) propose a research direction called 🤖agentic interpretability: we can and should ask and help AI systems to build mental models of us which will help us to build mental models of the LLMs. https://t.co/iw5lHnOlBU
0
0
0
Another yummy Ganguli gumbo! Nobody else can match him for: 1. Breadth of lines of evidence 2. Synthesis into insights about the inner workings of minds. Highly recommended!
Had fun presenting at @Harvard's @KempnerInst on Theories of rapid learning, creativity, and reasoning: of mice and machines Video: https://t.co/5Wzv9ZTToS Papers: Rapid learning: https://t.co/1zp8BSVUsy Creativity: https://t.co/a4JKHrR5U2 Math reasoning:
0
0
1
No one talks about the real reason driving the ~500k tech layoffs. Section 174 in the 2017 tax cuts turned engineer salaries from an instant tax deduction into a 5yr write off, causing billions in tax bills. It even incentivizes offshoring R&D, which has a 15yr write off! 1/4
216
1K
9K
A lot of people who think slow entry-level white-collar hiring is due to AI should take a look at this tax change and consider that there may be alternative causes.
8
21
172
A bit niche .... but this essay from pianist Jonathan Biss' and law prof Christopher Serkin about *interpretation* -- whether of music or law -- resonates with me. We have significant law *and* music from the 1780's, how do we bring them to life? https://t.co/QHmpYyr7Zf
nytimes.com
Classical musicians have a lot to teach interpreters of the U.S. Constitution. It’s so much more than the text.
0
1
0
Interesting thoughts from @signulll This UX shift has been shaping up for a while, but AI-first UX design will really accelerate it. Call it "Just in time UX"? Show just the info needed and in the format best suited to making it actionable.
i recently spent time with my six year old niece who now hates tapping on her tablet & it made me think that most people still don’t understand that the traditional human computer interaction stack already feels like a relic in so many ways. clicks, taps, swipes.. already feel
0
0
0
Computational Epistemologist* @rao2z at work! LLMs already have superhuman memory, and can assemble facts in plausible sounding ways. But we need stronger reasoning/planning to deliver real solutions using LLMs. So is it AGI? Not quite yet. *I made up the title. It's praise.
There is little doubt that LLMs--trained as they are on the collective digital footprints of humanity--"know" more than any single human. The AGI-level question is whether they know more than the humanity.
0
0
1