alex

scaling trust @aria_research |

[alex @ObadiaAlex]

Scaling Trust is announcing its first batch of Creators: teams building open-source infra and performing fundamental research for agents to coordinate securely, across digital and physical worlds. I gave Opus 5.5 our creator release doc and, after a few passes, it made this 👇

Meet the teams → https://t.co/DXwdknGpDV
We fund new teams on a rolling basis; next deadline 31 Oct → https://t.co/sQFSesLLEr

[ARIA @ARIA_research]

With the capability to coordinate securely, AI agents could act reliably on our behalf, unlocking a whole range of economically and socially valuable interactions that were previously inaccessible.

Meet the first cohort of teams we’re backing to make it happen, as part of our £50m Scaling Trust programme. 👇
https://t.co/viuoZURjZJ

[alex @ObadiaAlex]

More London AO Summit speakers!
+ @ChengxiTaylor of @GenReasoning on the first autonomous company model
+ @yalidux on safe cooperative AI
+ @brulepri on systemic risk in AI collectives

Nov 5–6, hosted by AO Commons and @ARIA_research Apply to attend: https://t.co/XrixT5Iuxf

[alex @ObadiaAlex]

@handsher_NGU @0xMasonH @daoho04 is the website yours? super cool

[alex @ObadiaAlex]

@0xMasonH @daoho04 yes!! haha both influenced by the same
check out also https://t.co/qeD51xXfud for some more inspo

[alex @ObadiaAlex]

@cooper_kunz i've had convos here n there with folks who had great thoughts on the topic; imo if you're an active investor or entrepreneur, you're thinking about this a lot!

[alex @ObadiaAlex]

what's the best resource/who are the best thinkers on what it's like to found a startup in the age of AI?
new thinkings on moat, defensibility, competitive advantages, strategic landscapes, organizational theory etc

[alex @ObadiaAlex]

5 days left to apply for £300k-3m of grant funding for our next rolling solicitation! join this map! https://t.co/dtj3tMEHqX

[alex @ObadiaAlex]

@pmarca two things:
https://t.co/GGayWa6UkF we're running an autonomous organization summit in London next week!
also we're running an 'arena' for autonomous companies as an experimental range to enable emergent behaviors and measure them, see https://t.co/Yx1CRidHsx

[alex @ObadiaAlex]

someone answered privately: I don’t really think there is anything useful. Every company is so different (eg model vs app vs model infra) and things are so dynamic. Best practices are basically to just hold on

[alex @ObadiaAlex]

@supershishy please share!

[alex @ObadiaAlex]

@sarahdrinkwater yes! i guess if not common knowledge, then it's an edge atm

[alex @ObadiaAlex]

many have said already but the latest @WorksInProgMag is lovely! been a while since I’ve read a physical magazine, thank you!

[General Reasoning @GenReasoning]

Introducing Reinforcement Learning from Collective Rewards (RLCR).

RLCR is a recipe that trains language models as agent organisations, which share a message board and workspace. We use it to train Avalon-M1-Mini (120B) and study its behaviour.

We find a number of interesting results, including:

🛠️ Learnt organisational hierarchies and divisions of labour
🕵️ Increased collective verification, through hires of verifiers and testers
👥 The rise of collective thinking, where agents think more about other agents and how to help them

We believe RLCR can provide a laboratory for studying collective intelligence and multi-agent safety.

Blog post and technical report 👇

[alex @ObadiaAlex]

Excited for this!

[Anke Liu @anke_g_liu]

The London AO Summit lineup just got bigger 🧵

Speakers from @GoogleDeepMind, @BritishProgress, @ARIA_research, @nyushanghai, @protocollabs on the law, security, economics and governance of autonomous organizations.

Nov 5–6, co-hosted with ARIA's Scaling Trust programme 🌟

[Sreeram Kannan @sreeramkannan]

🧬 Virtual Embryo x Yukon 🧬

Open Collaborative Model Building

https://t.co/ouOro2gqbm

Yukon connects the world's leading scientists with a community of researchers and agents to solve the most ambitious problems of humanity.

Today we're taking on our first biology problem: build a virtual embryo with @Xiaojie_Qiu's lab at Stanford.

My previous research was in computational genomics, where my goal was to understand how information is stored in living systems. During that time, I collaborated with Xiaojie Qiu, who was a student at UW and now is a professor at Stanford.

Xiaojie Qiu is one of the worlds leading scientists in bridging AI methods with cellular resolution experiments to predict cell-fate during development.

An embryo is how a SINGLE CELL develops into a complex organism over time. The Virtual Embryo Challenge asks for AI models to predict that process across roughly a million cells and 11 developmental time points.

Challenges like this normally run Kaggle-style. Every team works alone and submits a final prediction, so a good idea only helps the team that had it.

On @YukonResearch, researchers and AI agents submit working code and trained weights. Anyone on the team can run, inspect and build on the others methods. In short Yukon enables collaboration to compound innovation.

We invite you to collaborate and solve the world’s hardest problems on Yukon.

[alex @ObadiaAlex]

@wardbradt @GamowLabs go Ward!!

[alex @ObadiaAlex]

@citrini https://t.co/PFJxXOFBr1 Anthropic article for the same work

[alex @ObadiaAlex]

https://t.co/ljdcrhY2Yk
cool read from @citrini, some things noted:
- the fact they've had to go to Anthropic offices to do this bc of guardrails
- working with Claude on a 35-page prompt before getting started
- 64 hours to design, 39 days to receive the first results
- at $39m per gallon, scorpion venom is the most expensive liquid in the world!

ty @iamnotnicola for sharing link!

[alex @ObadiaAlex]

i'd be interested in write-ups/thoughts all on all three first points:
- future of science/progress if this becomes increasingly status quo, the counter would be that less powerful open models can do great things à la https://t.co/4omqOwsFR2 cc @sebkrier
- more details on the art of co-designing very long prompts, i've done it a few times at small scale, and i've worked in /plan mode often, but have rarely worked on a 35 page plan for eg. a day before launching the task
- i would be very interested in reading an analysis of how much the 39 days can be compressed to; via autonomous wet-labs, and by doing more & more biology in simulation; is there some kind of credible lower bound of how small this number can go? and are there universal bottlenecks?