- github.com/deepmind/meltingpot TODO vs DeepMind Lab2D? Also 2D discrete. Started in 2021.
- github.com/deepmind/ai-safety-gridworlds mentioned e.g. at www.youtube.com/watch?v=CGTkoUidQ8I by Rober Miles
Creating Multimodal Interactive Agents from DeepMind by Two Minute Papers (2023)
Source. www.deepmind.com/blog/building-interactive-agents-in-video-game-worldsOpen-Ended Learning Leads to Generally Capable Agents by DeepMind (2021)
Short name: XLand. Whitepaper: www.deepmind.com/blog/generally-capable-agents-emerge-from-open-ended-play.We define this category as AI games in which agents are able to produce or consume natural language.
It dawned on Ciro Santilli that it would be very difficult to classify an agent as an AGI if tthat agent can't speak to take orders, read existing human generated documentation, explain what it is doing, or ask for clarification.
Prototype: github.com/cirosantilli/Urho3D-cheat
Top Down 2D Continuous Game with Urho3D C++ SDL and Box2D for Reinforcement learning by Ciro Santilli (2018)
Source. Source code at: github.com/cirosantilli/Urho3D-cheat.Screenshot of the basketball stage of Ciro's 2D continuous game
. Source code at: github.com/cirosantilli/rl-game-2d-grid. Big kudos to game-icons.net for the sprites.Less good discrete prototype: github.com/cirosantilli/rl-game-2d-grid YouTube demo: Video 1. "Top Down 2D Continuous Game with Urho3D C++ SDL and Box2D for Reinforcement learning by Ciro Santilli (2018)".
Top Down 2D Discrete Tile Based Game with C++ SDL and Boost R-Tree for Reinforcement Learning by Ciro Santilli (2017)
Source. The goal of this project is to reach artificial general intelligence.
A few initiatives have created reasonable sets of robotics-like games for the purposes of AI development, most notably: OpenAI and DeepMind.
However, all projects so far have only created sets of unrelated games, or worse: focused on closed games designed for humans!
What is really needed is to create a single cohesive game world, designed specifically for this purpose, and with a very large number of game mechanics.
Notably, by "game mechanic" is meant "a magic aspect of the game world, which cannot be explained by object's location and inertia alone" in order to test the the missing link between continuous and discrete AI.
The question then becomes: do we have enough computational power to simulation a game worlds that is analogous enough to the real world, so that our AI algorithms will also apply to the real world?
To reduce computation requirements, it is better to focus on a 2D world at first. Such world with the right mechanics can break any AI, while still being faster to simulate than a 3D world.
The initial prototype uses the Urho3D open source game engine, and that is a reasonable project, but a raw Simple DirectMedia Layer + Box2D + OpenGL solution from scratch would be faster to develop for this use case, since Urho3D has a lot of human-gaming features that are not needed, and because 2019 Urho3D lead developers disagree with the China censored keyword attack.
Simulations such as these can be viewed as a form of synthetic data generation procedure, where the goal is to use computer worlds to reduce the costs of experiments and to improve reproducibility.
Ciro has always had a feeling that AI research in the 2020's is too unambitious. How many teams are actually aiming for AGI? When he then read Superintelligence by Nick Bostrom (2014) it said the same. AGI research has become a taboo in the early 21st century.
Related projects:
- github.com/deepmind/lab2d: 2D gridworld games, C++ with Lua bindings
Related ideas:
- www.youtube.com/watch?v=MHFrhIAj0ME?t=4183 Can't get you out of my head by Adam Curtis (2021) Part 1: Bloodshed on Wolf Mountain :)
- www.youtube.com/watch?v=EUjc1WuyPT8 AI alignment: Why It's Hard, and Where to Start by Eliezer Yudkowsky (2016)
Bibliograpy:
- agents.inf.ed.ac.uk/blog/multiagent-learning-environments/ Multi-Agent Learning Environments (2021) by Lukas Schäfer from the Autonomous agents research group of the University of Edinburgh. One of their games actually uses apples as visual represntation of rewards, exactly like Ciro's game. So funny. They also have a 2d continuous game: agents.inf.ed.ac.uk/blog/multiagent-learning-environments/#mpe
- humanoid robot simulation
- Section "AI training game"
- Section "Software-based artificial life"
OpenAI Plays Hide and Seek... and Breaks The Game! by Two Minute Papers (2019)
Source. Commentary of OpenAi's 2019 hide and seek paper. OpenAI does some similar simulations to what Ciro wants, but TODO do they publish source code for all of them? If not Ciro calls bullshit on non-reproducible research, and even worse due to the fake "Open" in the name. Does this repo contain everything?Much bigger simulation, AIs learn Phalanx by Pezzza's Work (2022)
Source. 2d agents with vision. Simple prey/predator scenario.github.com/deepmind/lab/tree/master/game_scripts/levels/contributed/dmlab30 has some good games with video demos on YouTube, though for some weird reason they are unlisted.
TODO get one of the games running. Instructions: github.com/deepmind/lab/blob/master/docs/users/build.md. This may helpgithub.com/deepmind/lab/issues/242: "Complete installation script for Ubuntu 20.04".
It is interesting how much overlap some of those have with Ciro's 2D reinforcement learning games
At twitter.com/togelius/status/1328404390114435072 called out on DeepMind Lab2D for not giving them credit on prior work!As seen from web.archive.org/web/20220331022932/http://gvgai.net/ though, DeepMind sponsored them at some point.
This very much looks like like GVGAI which was first released in 2014, been used in dozens (maybe hundreds) of papers, and for which one of the original developers was Tom Schaul at DeepMind...
www.gvgai.net (dead as of 2023)
The project kind of died circa 2020 it seems, a shame. Likely they funding ran out. The domain is dead as of 2023, last archive from 2022: web.archive.org/web/20220331022932/http://gvgai.net/. Marks as funded by DeepMind. Researchers really should use university/GitHub domain names!
Similar goals to Ciro's 2D reinforcement learning games, but they were focusing mostly on discrete games.
They have some source at: github.com/GAIGResearch/GVGAI TODO review
From QMUL Game AI Research Group:From other universities:TODO check:
- Simon M. Lucas: gaigresearch.github.io/members/Simon-Lucas, principal investigator
- Diego Perez Liebana www.linkedin.com/in/diegoperezliebana/
- Raluca D. Gaina: www.linkedin.com/in/raluca-gaina-347518114/ from Queen Mary
- Ahmed Khalifa
- Jialin Liu
2025 round one during week of April 21st, not toning down online profiles:Keywords:
- Paris quantum computing
- Alice&Bob:
- jobs.lever.co/alice-bob/2acf8f0c-6947-41ad-b30f-9779a1b49681 Senior Software Engineer, Framework team
I'm looking to do a meaningful job in a deeptech field, and quantum computing seems like it could become huge. I've learnt a few basics, and would like to go further with job experience in the area.
Salary: 90k.Rejected after first interview with the hiring manager. They said they wanted someone with more experience setting up CI/CD. - 2025-07-03 jobs.lever.co/alice-bob/b4632e27-cf56-4570-84bb-d56a169d1c43 Senior Software Engineer - Cloud. Webdev. I don't really want to do this, but it's a way in. Asking salary: 70k.
- Pasqal: careers.pasqal.com/jobs/5817098-software-development-engineer-integration Software Development Engineer IntegrationApplication auto-reply also points to:which is cool.I'm looking to do a meaningful job in a deeptech field, and quantum computing seems like it could become huge. I've learnt a few basics, and would like to go further with job experience in the area.Rejected 1 week later without interview even though my CV seemed like a perfect match for this job. Sent an email to the contributors of Pulser.After the application they keep sending you relevant new job openings, which was really good, applying to another one:I applied for this job careers.pasqal.com/jobs/5817098-software-development-engineer-integration but got rejected by HR lady without interview.Pinging you guys here just in case because sometimes my profiles scare the HR people and then when I ping the programmers they like me. I grabbed your emails from GitHub.
- 2025-06-20 careers.pasqal.com/jobs/6075845-embedded-software-engineer-w-m Embedded Software engineerRejected 2025-07-03.Why Pasqal?I want to get into quantum computing in France, and there are only 5 major hardware companies and I'm applying to each of them when there's a job that seems that I can do. I don't know enough to pick the one with the tech that is most promising to win, though it seems no one can at this point. Being in Massy, a little further from central Paris, is a plus, I like quieter areas.
- C12 Quantum Electronics
- 2025-06-20 Senior Measurement Software Engineer jobs.lever.co/c12qe/0b27ff8c-a094-4da4-ba32-2ecd55ae6233
I am interested in getting into quantum computing because the this field seems like it could be promising. My wife is French and working for a French company in Paris and we are looking to relocate to the greater Paris region. I've heard good things from the company. I have only done basic undergrad physics in university but I'm very interested in learning more from a more concrete point of view, AMO seems really cool. I can handle low level programming well.
- Welinq: none yet but add to loop
- Quandela: none yet but add to loop
- Alice&Bob:
- DeepMind:
- Turned down two days later before anything.As evidenced by my Stack Exchange contributions, I love trying out new software to see if it works and how well. I love benchmarking it. And I love documenting what I observed in great detail to help others to choose the best software for them. I also love meeting various new people and understanding what they are up to and how I can help.
- job-boards.greenhouse.io/deepmind/jobs/6576118 Software Engineer, Hardware Design
At ARM and AMD I've worked on various different parts of the hardware cycle, notably C++ emulation and low level software. I just get this strong feeling that hardware and low level software design should be the easiest thing for machines to automate, as it is the area that requires the least amount of "common sense" information about the world.
- Paris machine learning:
- H Company:Fastest initial job application steps ever! Name, email, CV, over.
- jobs.ashbyhq.com/hcompany/e6793ce6-918b-48a6-bf56-205c477cc1c0 Member of technical staff (Evaluations). Declined 12th June.
- jobs.ashbyhq.com/hcompany/89d867e7-2bd3-4918-aebc-cabbac526b6f Senior Back-end Engineer
- Mistral AI:
- jobs.lever.co/mistral/db67d7a2-bcec-4151-9b3a-8212ddabf419 Senior Software Engineer, Data Engineering - Paris. Declined May 7th without interview, at least they said they have lots of applicants and some have more closely related qualifications.
- 2025-07-03
- jobs.lever.co/mistral/c1e1e2c0-03f4-4434-938a-a5df3080baf2 Software Engineer, IDE (Typescript), (Contractor). VS Code browser extension.
Available from September. I made this extension as a small part of my personal project over a month full time: marketplace.visualstudio.com/items?itemName=ourbigbook.ourbigbook-vscode It has no installs because the personal project is a failure, but the extension works. I can do other stuff besides the vscode extension itself as well. But I need at least 1 year contract.
- I have 3 years full time on my failed website: ourbigbook.com/ My frontend is meh but backend is OK. Never done huge volumes though since the project was a failure. But Heroku + my basic PostgreSQL DB fu held up to 3k visits / day when I got some coverage for cirosantilli.com/cia-2010-covert-communication-websites from www.dailymail.co.uk/news/article-14752155/CIA-fake-websites-Star-Wars-communicate-spies.htmlI've also did some automatic population of the database to 100k rows and my indices seem to be right that the website is still usable: docs.ourbigbook.com/news/llm-generated-wikibot-abstracts
- jobs.lever.co/mistral/c1e1e2c0-03f4-4434-938a-a5df3080baf2 Software Engineer, IDE (Typescript), (Contractor). VS Code browser extension.
- Poolside AI:> Of all the applications of deep learning, code generation is one of those that interest me the most as they seem one of the most pertinent in order to one day achieve AGI (the others being theorem proving and robotics), and I'd like to try and get some work experience in the area, which is why I'm applying to your company.
- poolside.ai/careers/member-of-engineering-evaluations--ba11fe78-f6f6-4165-b76b-020a46ad8fee Member of Engineering (Evaluations)
Of all the applications of deep learning, code generation is one of those that interest me the most as they seem one of the most pertinent in order to one day achieve AGI (the others being theorem proving and robotics), and I'd like to try and get some work experience in the area.
- poolside.ai/careers/member-of-engineering-data-platform--13d32f62-d530-4372-b458-0687d99eea04 Member of Engineering (Data Platform)
This job seemed like a possibility as I've done some personal "data intensive" projects in the past (not distributed unfortunately, ~500 GB so it fit on my local disk), and I kind of enjoyed it and would be interested in trying out a more "data heavy" job like this for a change.
Rejected May 6th without interview. - poolside.ai/careers/member-of-engineering-evaluations--ba11fe78-f6f6-4165-b76b-020a46ad8fee Member of Engineering (Evaluations)
- H Company:
- Big tech:
- Google. Many jobs in London, nothing in Paris. www.google.com/about/careers/applications/jobs/results
- www.google.com/about/careers/applications/jobs/results/91141790538572486-senior-software-engineer-google-pixel-graphics Senior Software Engineer, Google Pixel Graphics
- www.google.com/about/careers/applications/jobs/results/96937277808091846-senior-software-engineer-pixel-graphics-gpu-software Senior Software Engineer, Pixel Graphics, GPU Software
- Amazon, focusing on AWS of course. Lots of jobs in London, nothing in Paris: www.amazon.jobs/en-gb/
- 2025-06-20
- www.amazon.jobs/en/jobs/2915117/software-development-engineer-aws-uk Software Development Engineer, AWS Utility Computing
- www.amazon.jobs/en-gb/jobs/2884321/systems-development-engineer-database-services Systems Development Engineer, Database Services. Rejected two days later on a Sunday, nice.
- www.amazon.jobs/en-gb/jobs/2530616/senior-systems-development-engineer Senior Systems Development Engineer
- www.amazon.jobs/en-gb/jobs/2902679/senior-systems-engineer Senior Systems Engineer
- 2025-06-20
- Microsoft: a few jobs in London, very few in Paris
- Facebook: several boring jobs in London, a few research jobs in Paris
- Google. Many jobs in London, nothing in Paris. www.google.com/about/careers/applications/jobs/results
- London machine learning:
- InstaDeep:
- www.instadeep.com/job-offer/fea583b3-d333-447e-8811-8ce58367c003/ Software Engineer (Simulation) asking 85k
I really like the idea of using simulations to speed up development of ideas or AI. I've done this in my past jobs in the semiconductor industry, and a bit on side time tinkering with simple simulation games that might be useful to train AI. I'm curious to what you will be simulating! The general areas that your company operates in, biotech and PCB design are also in my interest.lI really like the idea of using simulations to speed up development of ideas or AI. I've done this in my past jobs in the semiconductor industry, and a bit on side time tinkering with simple simulation games that might be useful to train AI. I'm curious to what you will be simulating!
- www.instadeep.com/job-offer/fea583b3-d333-447e-8811-8ce58367c003/ Software Engineer (Simulation) asking 85k
- Anthropic
- 2025-06-18
- job-boards.greenhouse.io/anthropic/jobs/4632138008 Software Engineer, Infrastructure. Declined first thing next Monday.
- job-boards.greenhouse.io/anthropic/jobs/4753256008 Software Engineer, Inference Scalability and Capability. This one suggested JAX, Rust and multi-GPU experience would be good. Haha it's hopeless.
- job-boards.greenhouse.io/anthropic/jobs/4632830008 Software Engineer, ML Performance and Scaling
I just want to help achieve AGI so I don't need to work anymore. I can do non-glamorous things to achieve that goal, but eventually I'd inevitably get into research itself. I haven't done deep learning except basic examples, but I can handle low level stuff, and I want to learn.
Why Anthropic?I'm applying to any LLM companies, and the ones with more funding are more likely to have dumb coding jobs that I can do without a PhD in ML, and more likely to have the resources to reach some critical point if scaling continues. The small labs have the advantage of nimbleness to try new architectures, but they only employ PhDs.In one paragraph, provide an example of something meaningful that you have done in line with your values. Examples could include past work, volunteering, civic engagement, community organizing, donations, family support, etc.*Endless open source knowledge sharing e.g.:- stackoverflow.com/users/895245 Stack Overflow
- github.com/cirosantilli/linux-kernel-module-cheat Linux kernel emulation tutorial
- ourbigbook.com/ my failed knowledge sharing website that only I use
- InstaDeep:
- Semiconductor:
- AMD:
* careers.amd.com/careers-home/jobs/57882 Senior Software Development Engineer on ROCm. Rejected a few days later without interview. - Qualcomm:
- Arm, all with referrals from old colleagues:
- 2025-06-30:
- GPU
- careers.arm.com/job/cambridge/staff-principal-software-engineer/33099/77726588832 Staff Software Engineer - System Performance Analysis GPU driver
- careers.arm.com/job/cambridge/staff-software-engineer-system-performance-analysis/33099/77284796288 Staff/Principal Software Engineer
- careers.arm.com/job/cambridge/principal-software-engineer-machine-learning/33099/82468207936 Principal Software Engineer - Machine Learning. Seems ambitions but friend recommended me so I just clicked. Rejected 2025-07-03.
- careers.arm.com/job/cambridge/staff-gpu-software-applications-engineer/33099/82643339264 Staff GPU software applications engineer. Got screening interview.
- Non-GPU
- carers.arm.com/job/cambridge/staff-software-engineer/33099/74848341808 Staff Software EngineerSounds cool.
This diverse role will largely focus on providing software enablement for our next generation CPU and accelerator based technologies on target use-cases.
- careers.arm.com/job/manchester/staff-application-performance-engineer/33099/80144718528 Staff Application Performance EngineerReally cool. I don't have the perf pre-requisites, but it would be cool to do this. Rejected 2025-07-03.
performance workflows on Arm Neoverse platforms
- careers.arm.com/job/cambridge/senior-software-engineer-machine-learning-tools/33099/74848310720 Senior Software Engineer - Machine Learning Tools. Got screening interview.
- experienced-arm.icims.com/jobs/14442/senior-linux-kernel-engineer-%e2%80%93-hardware-enablement Senior Linux Kernel Engineer – Hardware Enablement
- experienced-arm.icims.com/jobs/12987/staff-software-engineer/job?mode=submit_apply Staff Software Engineer
Accelerating algorithms with hand-optimized Arm assembly using the latest Arm technologies such as SVE
- carers.arm.com/job/cambridge/staff-software-engineer/33099/74848341808 Staff Software Engineer
- GPU
- 2025-06-30:
- Intel: no jobs in UK/France
- Nvidia: a few high end jobs in UK/France, some of them are EMEA remote
- 2025-06-30 Senior Software and System Architect nvidia.wd5.myworkdayjobs.com/en-US/NVIDIAExternalCareerSite/job/Spain-Remote/Senior-Software-and-System-Architect_JR1997578?locationHierarchy1=2fcb99c455831013ea52aa2df70e324e Read wrong, seems to require senior networking skills that I don't have, but by then had already asked for a referral from an ex colleague and so just applied anyways. Lol!
- Codasip: no SW jobs in UK, only HW
- SiFive: only 10 jobs globally, 1 sales in UK
- Graphcore:
- 2025-07-02 job-boards.greenhouse.io/graphcore/jobs/7743036002 Staff System Software Engineer. Automation and testing stuff.
- AMD:
- Open source
- Canonical
- Linux kernel engineering: failed after take home exercise
- Common Crawl: commoncrawl.org/jobs Software Engineer/Data Scientist, Python, Spark
- Canonical
- Random compute
- Atos, now mostly Eviden: all jobs require 10 years of experience lol
- Random french tech
- Dassault Systemes
- Thales
- biotech maybe
- optical compute
- NcodiN: no jobs
- Flux Computing:
- robotics
- Enchanted Tools
- 2025-06-26 taleez.com/apply/core-c-developer-11e-arrondissement-paris-enchanted-tools-cdi Core C++ developerSalary: 80k.
Just want to help achieve AGI and do some cool programming.
- 2025-06-26 taleez.com/apply/core-c-developer-11e-arrondissement-paris-enchanted-tools-cdi Core C++ developer
- Enchanted Tools
- HPC
- GPU
- C++
- Linux
- systems software
- numerical software
- robotics
Was a closed source project by "Roboti LLC", which was then acquired by DeepMind in October 2021 and open sourced March 2022: www.deepmind.com/blog/open-sourcing-mujoco
Tested on Ubuntu 23.10;
git clone https://github.com/google-deepmind/mujoco
cd mujoco
git checkout 5d46c39529819d1b31249e249ca399f306a108ac
mkdir -p build
cd build
cmake ..
make -j
Now let's play. Minimal interactive UI simulation of a simple MJCF scene with one falling cube:Test soure code: github.com/google-deepmind/mujoco/blob/5d46c39529819d1b31249e249ca399f306a108ac/sample/basic.cc. The only thing you can do is rotate the scene with the computer mouse it seems. Mentioned at: mujoco.readthedocs.io/en/2.2.2/programming.html#sabasic
bin/basic ../doc/_static/hello.xml
Some more interesting models can be found under the
model/
directory: github.com/google-deepmind/mujoco/tree/5d46c39529819d1b31249e249ca399f306a108ac/model E.g. the imaginary humanoid robot DeepMind used in many demos can be seen with:bin/basic ../model/humanoid/humanoid.xml
A more advanced UI with a few controls:Test soure code: github.com/google-deepmind/mujoco/tree/5d46c39529819d1b31249e249ca399f306a108ac/simulate. Mentioned at: mujoco.readthedocs.io/en/2.2.2/programming.html#sasimulate
bin/simulate ../doc/_static/hello.xml
A very cool thing about that UI is that you can manually control joints. There are no joints in the hello.xml, but e.g. with the humanoid model:under "Control" you move each joint of the robot separately which is quite cool.
bin/simulate ../model/humanoid/humanoid.xml
There's also a Mentioned at: mujoco.readthedocs.io/en/2.2.2/programming.html#sarecord but TODO that produced a broken video, related issues:
bin/record
test executable that presumably renders the simulation directly to a file:bin/record ../doc/_static/hello.xml 5 60 rgb.out
ffmpeg -f rawvideo -pixel_format rgb24 -video_size 800x800 -framerate 60 -i rgb.out -vf "vflip" video.mp4
DeepMind likes coming up with new improved algorithms for these more specific cases, e.g. it was announced in 2025 that AlphaEvolve found a novel 4x4 complex valued algorithm that uses 48 multiplications.
Bibliography:
- fmm.univ-lille.fr/ attempts to keep an up-to-date list for various sizes
Updates Understanding the state of 3x3 matrix multiplication Updated 2025-07-01 +Created 2025-05-21
After yet another awesome announcement by DeepMind that it had improved theoretical 4x4 matrix multiplication reducing the number of scalar multiplications with its AlphaEvolve system, I decided to have a look at the smallest open size 3x3 to understand what was going on in there.