Tao Burga
@taoburr
Followers
1K
Following
7K
Media
146
Statuses
756
Technology Fellow @IFP | AI+science, compute policy, and def/acc R&D. | Prev. @NYUniversity researcher, @iapsAI Fellow, @BrownUniversity.
Washington, DC
Joined October 2015
So proud of this, and grateful to all the authors & IFP team for pulling it together! . Go check it out š.
THE LAUNCH SEQUENCE. A new collection of 16 essays on how to accelerate AI for science & security: The AI revolution is already delivering enormous consumer benefits. But AI progress won't automatically solve humanity's most important problems first. To
1
7
47
(2/2) I don't think this is particularly informative as to the "intentions" of the models, but we should be benchmarking these propensities against human experts if we'll integrate LLMs into military systems.
0
0
1
The most fun image & video creation tool in the world is here. Try it for free in the Grok App.
0
215
1K
Yep -- and in my own research I found some LLMs to be more escalatory than humans in simulated nuclear crises. I made a tiny benchmark for nuclear risk perception, aggression, & cascading failure from multi-LLM decision support. I should maybe re-run the study with newer LLMs.
Five off-the-shelf LLMs played with war games. "Almost all of the AI models showed a preference to escalate aggressively, use firepower indiscriminately and turn crises into shooting wars ā even to the point of launching nuclear weapons.". You don't say!.
1
1
3
Since Nick Land was namedropped so much, I'll just leave this here. "Machinic desire" and "hyperstition" are top-tier concepts.
0
1
16
Many people are saying.
0
0
5
RT @RyanPGreenblatt: I'm skeptical of claims that some specific advance will cause very above trend AI progress in the next year. Ongoing bā¦.
0
6
0
RT @colemcfaul: šØ NEW REPORT! šØ. After 2 (!!) years working on a new dataset of thousands of PLA contracts, Iām very excited to share a newā¦.
0
161
0
RT @nickgillespie: More to the mix: People are generally happy with their personal situations but are bothered by everyone else's. Excellenā¦.
0
70
0
RT @calebwatney: LLMs scaled rapidly because the internet gave us oceans of structured language data. But we lack equivalent datasets forā¦.
ifp.org
How block-grant labs can generate the real-world data AI needs to do science
0
21
0
RT @ARIA_research: ARIA is launching a multi-phased solicitation to develop a general-purpose Safeguarded AI workflow, backed by up to Ā£19mā¦.
0
7
0
RT @AlecStapp: Reminder: one of China's biggest fears is that someday we might actually fix our broken immigration system (instead of activā¦.
0
104
0
As the stakes increase, AI labs will seek more and more security for their IP.
0
0
4
RT @BigMeanInternet: Every single comment is like "wow! how much faster does it go š" and it simply does not.
0
4K
0
Okay now letās do actual high-speed rail to convert DC-Baltimore-Philly-NYC-Boston into the One Big Beautiful Northeastern Megalopolis
And that marks the first ever Boston ā D.C. journey on @Amtrakās NextGen Acela and a new era of passenger rail in the U.S.!
2
2
26
RT @BenSchifman: In my house growing up we had a trophy my mom won in a high school speech tournament. Her topic? Zero population growth.ā¦.
0
6
0
Idk if Iām just not getting how to use GPT 5 well yet but I find it unusable. Even with manual switching, I just prefer Opus/Sonnet for big and small tasks.
2
0
6
RT @NiskanenCenter: NEW PAPER: The emerging āAbundance movementā isnāt left, right, or centerāitās a cross-cutting idea. Here are six difā¦.
0
106
0
RT @peterwildeford: Wow who knew this data center thing would be such a big deal?.
0
2
0