Let's at least ensure that: if citations can be nouns, the citation format must be more self-identifying than, e.g., [21]. Purely numeric citations are the standard in a lot of ACM venues and they noun very poorly.
@daveandersen
Computer Science Professor, CMU; co-founder and CTO, Enriched Ag Energy-efficient computing, a dash of security, and a pinch of databases. Also on Mastodon (https://hachyderm.io/@dave_andersen ) signal: dga.48 he/him
Let's at least ensure that: if citations can be nouns, the citation format must be more self-identifying than, e.g., [21]. Purely numeric citations are the standard in a lot of ACM venues and they noun very poorly.
A big pothole on Forbes Ave is exposing the old streetcar trolley tracks from #pittsburgh past.
(The black and white photo is from retrographer of the 69 Squirrel Hill trolley)
I feel out of touch. Was talking with a colleague who runs supercomputers at a national lab who mentioned casually that they have individual racks at 1/2 MW each. (!)
mind-boggling power density. The water cooling pipes alone are huge.
The "My name is Ayatollah Khamenei, you killed my father" quips just write themselves.
If only Stephen Miller had six fingers.
It turns out to be possible to have nice things with appropriate investment. Pittsburgh in 2045 will have lower crime rates than it would without this. Long time horizon but worthwhile.
Made okonomiyaki tonight. Nom.
Maybe! That's really more of a human question about how they deal with something that seems authoritative but has errors hiding in it. I don't know the answer there. I'm sure there are places in the workflow they could be useful, but suspect integrating is hard work in collab with domain experts
This makes half of my brain explode in a good way and half of it wonder about correctness and how to validate it.
Also I love the json interface. I wish more tools did that. I'm sure powershell fans are laughing at me now but I didn't know powershell. :)
This is a very nice explanation of why LLMs should be viewed skeptically for both law and medicine. You don't get to run your brief or diagnosis through a battery of automated, non-erroneous tests.
LLMs work when the cost of validation is much less than the cost of production (+ output likely good)
shit
This whole thing has a lot of "let's just hope the random dice roll of chaos works out in our favor" energy, which sounds like the kind of strategy a bunch of people might adopt if their worldview is that thinking carefully is hard, painful, and should be avoided at all costs.
(The PDL does not primarily focus on formal verification, but the trend lines make it clear that verification is going to play an increasing role in mission-critical systems development over time, so we're roping in Bryan to add a dash of "big things you should keep a finger on" to our mix.)
Monday's market open is going to be a doozy
Also, it turns out there's a real danger of becoming the co-director of the lab: There are fewer grown ups in a position to tell me that my jokes aren't funny. One of my slides from the intro keynote:
We're holding the big lab+industry retreat this week, and I'm tremendously excited that we've been able to twist Bryan Parno's arm into giving a longer talk on "Verification In The Age of Vibe Coding". π€
war is peace
ignorance is strength
oh, so it's concepts of a SCIF
Time for a better LLM
Preach it, Abbey
Former openai policy lead
One consequence of attempting to make a plan is that you might learn that military force is unlikely to achieve your objectives.
This is the problem of electing an administration of sycophantic toadies to a demented orange blob. There are lots of folks in DoD who could point out all the things missing (like, say, "an actual plan"). But they're ignored or get fired. B player hire C players, as they say, and these aren't even B
Dean W. Ball Y @deanwball X.com Think about the power Hegseth is asserting here. He is claiming that the DoD can force all contractors to stop doing business of any kind with arbitrary other companies. In other words, every operating system vendor, every manufacturer of hardware, every hyperscaler, every type of firm the DoD contracts withβall their services and products can be denied to any economic actor at will by the Secretary of War. This is obviously a psychotic power grab. It is almost surely illegal, but the message it sends is that the United States Government is a completely unreliable partner for any kind of business. The damage done to our business environment is profound. No amount of deregulatory vibes sent by this administration matters compared to this arson.
Scary af, by @deanwb.bsky.social
Microsoft does this:
learn.microsoft.com/en-us/copilo...
learn.microsoft.com/en-us/copilo...
They claim, e.g., both HIPAA and ferpa compliance for queries and responses under the enterprise data protection option.
Someone: "Hey, wait, the US seems to be doing pretty well on AI, bubble or not, it's holding the economy up"
Hegseth: "Hold my beer"
(just kidding, he kept the beer to drink with kesh patel when they celebrate the demise of the US economy)
Dario's statement literally just says "hey, no fully autonomous weapons and no use for mass surveillance of us persons". The reactions are ridiculously out of range for what are very reasonable boundaries. It's a bad idea to stock the DoD with children who throw a tantrum when told no.
MAHA?
Nuh-uh.
You won't hear one croak from RFK, Jr. on this.
Abusers react poorly to people establishing boundaries, exhibit 5001.
Good for Anthropic for being clear about their boundary. Anyone who's ever interacted with an LLM understands why a fully autonomous weapon is a terrible idea now
"My bad, you're absolutely right! That _was_ a bus full of nuns!"
please make the scary bad thing go away.
I think it was intended rhetorically. "What does he care?" to my ear has a strong implication of "he doesn't". I wonder if this is yet another regionalism I didn't realize was in my head. π€