Bryan Bischof fka Dr. Donut (@BEBischof) header
Bryan Bischof fka Dr. Donut (@BEBischof) profile picture

Leading AI @theoryvc; Teach ML @rutgersu. Prev: Led AI @_hex_tech; Led Data @weights_biases, ML @stitchfix, Led Data @bluebottleroast

Ohlone Territory
Joined December 2010
Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)Photo by Bryan Bischof fka Dr. Donut (@BEBischof)
A rat can run a restaurant, but can an agent run a bakery? Welcome to The Great Business Bake Off: our AI Agent Hackathon. On December 11, you and your agents get five hours to make the most profitable bakery in town. Spots are limited: theoryvc.com/bakeoff
42
30
31
137
41,151
Finite geometry was my first love in mathematics
Latest newsletter: Finite geometries Projective, affine, and hyperbolic johndcookconsulting.substack…
3
305
Can we get that account that makes song explainers on this?
Norway now has more than 2,000 underwater AI laser robots shooting lice off its salmon they're made by a company called Stingray, and they already guard more than 50 million salmon and trout 24/7 the reason is sea lice, tiny parasites that latch onto salmon and feed on them as few as 11 can kill a small fish, and one pen can hold 150,000 salmon the usual fix is pumping the fish out and blasting them with warm water or water jets to knock the lice off but the handling is so rough that Norway's fish health report ranks it the top cause of death in full-grown farmed salmon so instead, Stingray puts a robot inside the pen that hunts lice one at a time, all day: > two cameras watch every salmon that swims past > AI checks two frames to make sure it's really a louse > then a green laser fires from up to 2 meters away > the salmon's scales reflect the beam, so the fish swims off fine and the louse is dead we now live in a world where AI lasers do pest control on your dinner lmao
1
336
There are approximately infinite use cases for this.
Want to do AI-powered data extraction 2x faster and cheaper than other carefully-engineered inference solutions? Check our new AI-powered extraction technique in Quail, our open-source query engine with AI functions!
2
16
3,861
Some of my coworkers are now calling it Chad instead of Chat and boy does this change the vibe
4
9
1,785
A mascot has dropped in chat:
101
It's giving creepy ddr
full cut: "one last breath sung" by AI, arrangement unknown, edited by yours truly
1
4
623
This is actually so good
1
2
82
Bryan Bischof fka Dr. Donut retweeted
Theory Ventures AI head Bryan Bischof, whose team runs thousands of agent jobs, on why the GPU shortage isn't universal: "Maybe five to 10% is right now." He'd like another 40–60% before his next meeting. The rest can wait until sometime that week. "Not all jobs require right now, and not all jobs require frontier intelligence either." "When people say, 'I just can't get the GPUs I need,' we'll ask them, 'What are you doing?'" A trillion-parameter training run may need a thousand B300 GPUs available for researchers to schedule immediately. A small agent summarizing text is a different compute request. "There's something out there for you. I promise." @bebischof
2
3
2
807
This is sick. I love it
My former Google colleague Matthieu and I have launched our new recipe and cooking app, apinch! It's a recipe manager where you can follow what your friends are cooking – a Strava for home cooks. Load old recipe cards; add what you see on Insta; record what works for you!
3
499
Excellent project idea; terrifying implications
🔥 NEW PROJECT🔥 I tried to find Wally/Waldo lookalikes on unsecured CCTV cameras. The first results came in after a few days, so I let the GPU server run for weeks. You can find more information and results on this page: wallyfinder.com My projects: driesdepoorter.be/
7
429
This is cool
𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝗦𝗶𝗱𝗲𝗾𝘂𝗲𝘀𝘁 🤑 Sidequest is a Claude Code mod that pays developers to see ads while they work. → One short “Sponsored” line appears under Claude’s response → Developers earn from every ad they see and get paid our through Stripe → Each ad is matched to your session and never changes Claude’s output
3
2
7
773
This is exciting. I’ve wanted to work on something like this for myself so excited someone has
Introducing Lattice, a local LaTeX workspace for your research. A paper is more than a .tex file: the papers you cite, notes, slides, Overleaf co-authors and now an AI agent. I wanted it all in one app, so I built it. Free to download, open source: leo1oel.github.io/lattice
7
731
Bryan Bischof fka Dr. Donut retweeted
Replying to @BEBischof
xitter.cf/pvncher/status/2113794… but /goal is also apparently an antipattern?? lol why is it so hard to do the right thing with OpenAI models + Codex?
Replying to @willdepue
Hey I really recommend not using goal. You’re much better off with a manager agent paired with a scheduled automation that reminds it to poll progress of different runner agents it starts in separate threads. My pro tip after that, is to tell the manager to write all your steers down, and reconcile them against the overall goal before dispatching new work.
1
1
3
285
Got a lot of great responses both in thread and dm. Allow me to summarize: - my experience is widespread and a lot of people are frustrated. - many people have switched to Claude over it and opus5.5 is very good. Apparently Claude app is improved from where it used to be terrible - /goal is a way people are working around this in codex when possible Cc @OpenAIDevs
Before Astra came out, codex + sol medium (and a little high) were quite good at continuing to execute. I would do large projects with many steps, and I could get them to: - do a pretty good job of planning - take meaty steps that actually made progress - ask the right followups to stay on track. Astra broke this completely, on any thinking level astra simply will not continue working and rarely sets up the project in a way that it can iterate through a lot of work. Similarly, 6.1 sol has the same problem, but even more stop-and-go. Even more 'i'm sorry, i explained it but didnt do it' or 'next is X' and then nothing. Before 6, I was thinking Codex was fully taking over as my go-to harness because it was something throughout the day i could poke and keep getting more work done. Now, days go by with virtually no progress. Has anyone migrated off Codex+remote in a way that preserved their context well enough and has good mobile support?
3
17
1,325
Bryan Bischof fka Dr. Donut retweeted
A rat can run a restaurant, but can an agent run a bakery? Welcome to The Great Business Bake Off: our AI Agent Hackathon. On December 11, you and your agents get five hours to make the most profitable bakery in town. Spots are limited: theoryvc.com/bakeoff
42
30
31
137
41,151
constant
Before Astra came out, codex + sol medium (and a little high) were quite good at continuing to execute. I would do large projects with many steps, and I could get them to: - do a pretty good job of planning - take meaty steps that actually made progress - ask the right followups to stay on track. Astra broke this completely, on any thinking level astra simply will not continue working and rarely sets up the project in a way that it can iterate through a lot of work. Similarly, 6.1 sol has the same problem, but even more stop-and-go. Even more 'i'm sorry, i explained it but didnt do it' or 'next is X' and then nothing. Before 6, I was thinking Codex was fully taking over as my go-to harness because it was something throughout the day i could poke and keep getting more work done. Now, days go by with virtually no progress. Has anyone migrated off Codex+remote in a way that preserved their context well enough and has good mobile support?
9
18
1,541
I really shouldn't have complained, now it's worse!
1
2
194
@OpenAIDevs guess this is the last time i use y'alls containers
2
153
Before Astra came out, codex + sol medium (and a little high) were quite good at continuing to execute. I would do large projects with many steps, and I could get them to: - do a pretty good job of planning - take meaty steps that actually made progress - ask the right followups to stay on track. Astra broke this completely, on any thinking level astra simply will not continue working and rarely sets up the project in a way that it can iterate through a lot of work. Similarly, 6.1 sol has the same problem, but even more stop-and-go. Even more 'i'm sorry, i explained it but didnt do it' or 'next is X' and then nothing. Before 6, I was thinking Codex was fully taking over as my go-to harness because it was something throughout the day i could poke and keep getting more work done. Now, days go by with virtually no progress. Has anyone migrated off Codex+remote in a way that preserved their context well enough and has good mobile support?
36
2
1
48
9,839
Bryan Bischof fka Dr. Donut retweeted
I love this. We need a lot more of it. This is how AI actually reaches people who don't follow any of this. The format is cool and familiar. People need to see what agents can do, and this helps a lot. Congrats, @BEBischof 👊🏼
A rat can run a restaurant, but can an agent run a bakery? Welcome to The Great Business Bake Off: our AI Agent Hackathon. On December 11, you and your agents get five hours to make the most profitable bakery in town. Spots are limited: theoryvc.com/bakeoff
1
2
259