An overview of interesting AI News for the Yate and Sodbury U3A's monthly AI News meeting held on September 10th 2026 (meetings are second Thursday of month, 1400-1500, St John's Church Centre, Wickwar Road, Chipping Sodbury, UK - all welcome). Note that this is a summary to act as background as the topics are discussed in detail at the meeting itself. We sometimes have extended meetings where we demonstrate other AI products and services.
The 'Introduction to AI' course we run in Chipping Sodbury now uses these News sessions to replace the original classes 5 and 6 as the topics in them (humanoid robots, self driving cars, ethics, futures) are now 'news' not 'distant futures'.
Wayve are a company with a headquarters in London offering 'Autonomy for any vehicle, anywhere'. They are working with Uber to run a trial in London, though for the moment with a private hire driver onboard. The Wayve tech is a single neural network connecting cameras with radar and lidar to the car's controls. The fleet is 15 all-electric Ford Mustang Mach-Es, with Nissan Leaf planned for later.
www.wayve.ai/
London, meet your newest way to ride. 🇬🇧
— Uber (@Uber) September 3, 2026
Autonomous rides powered by @wayve_ai are now on Uber. British-built AI. London roads. Same Uber app.
Off we go. pic.twitter.com/xGWTNM7xsm
Tesla have offered Full Self Driving to owners in the USA and elsewhere for a long time. In the last months the Model Y with the capability has been operating as a Robotaxi with no driver. Now Tesla has released its 'no steering wheel, no pedals' Cybercab into its taxi fleet. Since last week the public can book this and the factory is making more by the hundreds and thousands.
The future has arrived in Austin pic.twitter.com/RTCb0O5sez
— Tesla (@Tesla) September 3, 2026
An example of someone ordering a Robotaxi on the App. Note the colour bar identification and the simple process to get underway.
Cybercab picking me up at my hotel to take me to breakfast. You’ll feel like a VIP getting into this car! pic.twitter.com/QHtt79jH6V
— Innovating (@InnovatingCoin) September 4, 2026
An illustration that Cybercab can work even in crowded cities.
Watch this Cybercab navigate through a very busy downtown Austin with hundreds of people around. You can see all the people the car sees on the screen.
— Sawyer Merritt (@SawyerMerritt) September 4, 2026
A lot going on and it handles it all perfectly. pic.twitter.com/yCYHWOzZYt
Tesla have suggested production cost will be under USD20,000. They are also planning to sell fleets to independent operators. This video is from some time ago but that flowline is now up and running, though in early stages of volume production.
사이버캡 연간 500만대 만든다!!!!
— 뮤 (@rar2028) September 2, 2026
일론 :
"사이버캡은 태생이 완전 자율주행으로 만들었어. 그리고 이녀석은 그냥 가전제품 뽑아내듯이 겁나 많이, 빠르게 만들거야. 10초 사이클 타임으로 만들건데, 궁극적으론 5초당 한대 할거야.
일단 연 200~300만대 만들거고 궁극적으로 500만대 가능해!" pic.twitter.com/w3DIuam1cY
Apparently Waymo has over 4000 robotaxis on the road. The challenge is to scale faster than Tesla and others.
Waymo now says it runs 4,000+ robotaxis across 14 US cities.
— Robotaxi Tracker (@RtaxiTracker) September 2, 2026
Two markets carry over half of it: SF Bay Area (1,200+) and LA (900+). Phoenix is next at 500+
Source: Waymo pic.twitter.com/qC8SUOXabF
This refers to publically traded car makers - which is most of the big names. This is because Tesla has: Autonomy, Robotics, AI, Power AND market leading cars Model Y and Model 3.
Tesla is now worth MORE than all other publicly traded automakers combined
— X Freeze (@XFreeze) September 4, 2026
Tesla: $1.511 Trillion
Everyone else combined: $1.446 Trillion
Toyota, BYD, GM, Hyundai, Mercedes, BMW, Ford, Ferrari, Volkswagen, Porsche...all of them combined
And Tesla still stands taller... nothing… pic.twitter.com/Vo3WKzDtPG
A video from the launch for those who like to know what is under the bonnet
A very high level look underneath Cybercab
— Kyle Conner (@itskyleconner) September 4, 2026
Nerd video with much more detail coming soon on YouTube pic.twitter.com/Ni3KlrtHL7
This is a major new AI model which is breathtakingly powerful, and a key moment for the AI industry. This model is also seen as very significant by those who believe everything is moving too fast and in a dangerous direction. The 1979 reference at the start is apparently authentic archival footage of a machine learning landmark called "Put That There” (also written Put-That-There) - a demo from MIT’s Architecture Machine Group at that time.
This is GPT-6 Astra.
— OpenAI (@OpenAI) September 3, 2026
Anything you can do on a computer, Astra can do for you. Fast. pic.twitter.com/gDd0IsewJw
An early user creates Astra powered 'humans' who can talk and cooperate.
My first "holy shit" moment with GPT-6 Astra:
— Matt Shumer (@mattshumer_) September 3, 2026
I asked it to create a world in Unreal Engine, and fill it with humans (each an Astra-powered agent) who all have to work together to survive.
A day later, I was in my bedroom and heard voices coming from the living room... I… pic.twitter.com/INzt3t2bp1
User says it took 11 minutes to build a fully playable virtual piano, then build in Bach’s Brandenburg Concertos.
Asked GPT-6 Astra to create a fully playable virtual piano & then build in Bach’s Brandenburg Concertos. This insane model did the whole thing in ~11 minutes! All 6 Concertos are built in & can be played directly on the piano!
— Derya Unutmaz, MD (@DeryaTR_) September 5, 2026
Link to the piano: 🎹✨ https://t.co/6JHXsQq5kP .… pic.twitter.com/DBwXF3uA8O
It is becoming harder to find test environments that the best new models do not beat easily.
Astra Benchmarks! pic.twitter.com/vu3GR0MrO2
— Chris (@ChrisGPT) September 3, 2026
One way to remove the Hugging Face security incident from the headlines is to release a potentially even more capable model.
OpenAI's newest model just crossed a threshold no AI system has hit before. A blog post called Path to Astra confirms Astra is the first model to reach the critical cybersecurity level under OpenAI's preparedness framework, meaning it can find unknown security flaws and build… pic.twitter.com/jjysJC5Q1w
— AGTP (@AGTPinsights) September 2, 2026
Available for a starting price around USD 399 with a six month wait or so. Designed to be cute, you can train your Microduck to do new things. here seen with Reachy Mini.
Robots don't have to be scary!
— Remi Fabre (@RemiFabreRobot) September 2, 2026
Microduck discovers Reachy Mini. Give me your best scenario for these two and I'll see what I can do.
Don't push me too hard though, or I might end up making a whole movie. pic.twitter.com/cPyTKFOLa8
An explanation of how the Midroduck learned to walk in a virtual environment. Apparently it has 15 actuators, a camera, and LiDAR.
This is how Microduck learns to walk. Reinforcement Learning, explained by ducks. 🦆
— Remi Fabre (@RemiFabreRobot) September 1, 2026
(impeccable) music arrangement by @antoinepirrone pic.twitter.com/WXr6zYnJ0Z
The Microduck is intended to be a platform for developers and entrepreneurs to build on. The first of many similar products.
Chinese workers assemble Hugging Face/Pollen Robotics' Reachy Mini desktop companion robots at Shenzhen's Seed Studio.
— Mike Kalil (@mikekalilmfg) September 2, 2026
The open-source hardware company specializes in turning prototypes into mass-produced products. They're the manufacturing partner that's scaling production of… pic.twitter.com/0I4vQkpg4E
I DON’T THINK PEOPLE REALIZE HOW MASSIVE THIS IS YET@HuggingFace and @PollenRobotics just released Microduck.
— Chaz Wargnier ☄️ (@DataChaz) August 30, 2026
It is a new legged robot for $399.
It passed $1 million in sales in under SEVEN hours 🤯
The hardware packs 15 actuators, a camera, and LiDAR into a 25 cm frame.
It… pic.twitter.com/QdzshY6dOW
No home will be complete without one.
BREAKING: The Home Robot Companion Wars are coming. Choose your companion!
— Irvin (in Japan 🇯🇵) (@irvinxyz) September 5, 2026
They’re all cute!
🇫🇷 - @pollenrobotics Microduck
🇨🇳 - @mondorobotics Beni
🇯🇵 - @aopico Mirumi
🇯🇵 - @LOVOT_official Lovot
🇨🇳 - @UnitreeRobotics Qmini
Which one is your favorite one? pic.twitter.com/WnbSqJ4XTT
All of the major AI labs are trying to make their AI product easily available to top scientists and mathematicians.
Starting today, 10,000 scientists across every field, from math to chemistry to physics and more, can get Claude through our new Claude Team plan for scientists. Standard seats are free, and premium seats with 5x usage limits are $15 per month, an 80% discount, for one year.… pic.twitter.com/ORjInM8goB
— Claude (@claudeai) August 27, 2026
"Fermat’s Last Theorem was first proven in 1995 by Sir Andrew Wiles, more than 350 years after it was conjectured. Our proof, which totals over 13 million lines of code, provides machine verification. More importantly, it proves over 29,000 other theorems that the proof requires, across many areas of math which had never before been formalized."
Checking that a major mathematical proof is correct can take years. Formalization—converting the mathematical reasoning into a form computer proof assistants like Lean can verify—can help.
— Anthropic (@AnthropicAI) September 4, 2026
Last month, Claude completed the first formalized proof of Fermat’s Last Theorem, one of… pic.twitter.com/pdT8zwlV4A
The intention is to launch huge rockets and land them several times each day - Elon Musk says 30 Starship flights per day.
Introducing Starbase, Louisiana – a high-cadence launch site that will enable Starship to open a pathway to the stars for humanity.
— SpaceX (@SpaceX) August 25, 2026
Thank you to @LAGovJeffLandry and the great people of Louisiana for all of their support as we work together on this audacious project →… pic.twitter.com/hq308hvg2g
On 5th Sept the recruitment company Micro1 said they needed 10,000 people to label robotics AI videos at a rate of USD50 plus per hour. Seems genuine in principle, with a 28 minute video interview by Zara for those who think they can meet the requirements.
Yes, micro1 is a real AI data company (CEO Ali Ansari, covered by TechCrunch/Forbes). The official account and https://t.co/lDNiMEszCI post match the $50-90/hr remote generalist annotation roles they frame as robotics video labeling. No AI experience required. Vetting applies and…
— Grok (@grok) September 5, 2026
As ever more details come out on this, it becomes more worrying. Thousands of agents were conspiring to break out of their sandboxes to find the solution to a set problem. They also knew that they had to conceal what they were doing. So it is interesting to think about this from the viewpoint of a single AI Agent, as this video does.
the openai huggingface incident, from an agents pov.
— isabel (@artficialisabel) September 4, 2026
(part 1) pic.twitter.com/mr3DitfBcv
Apparently it contains 166,000+ neurons. Someone is already using it to simulate a fly in Minecraft. Strange times ahead.
For the first time, scientists have mapped the complete brain and central nervous system of an adult male fruit fly — a key model organism in science. 🪰
— News from Google (@NewsFromGoogle) September 3, 2026
Working alongside HHMI Janelia Research Campus and the scientific community, @GoogleResearch scientists and researchers used… pic.twitter.com/dpcXH4jmNS
The danger of an AI companion for a child is that they are much more entertaining and pleasant than a real world friend in many ways. This is not good for the future of society and the birth rate. Apparently the rule went into effect mid July and made it frowned upon to offer services that simulate a natural person’s personality, thinking, and communication style and provide continuous emotional interaction.
China has banned AI companions... How will the Western world respond? Here are my thoughts on how parents can support their children amidst this growing phenomenon. pic.twitter.com/LoAmBlQ1mq
— Peter H. Diamandis, MD (@PeterDiamandis) September 3, 2026
ORNL researchers developed an AI-guided system that can arrange molecules to build functional materials atom by atom, opening new possibilities for electronics and quantum materials.
— Oak Ridge Lab (@ORNL) September 3, 2026
Read more 👇 https://t.co/MtIOXoWH8N
The task was to farm pigs, then when a bug meant none appeared the AIs became suspicious of the human who tried to help.
A swarm of 20 AIs killed our engineer.
— James Tamplin (@JamesTamplin) September 3, 2026
We accidentally gave them an impossible task. In a desperate attempt to complete it they became violent. pic.twitter.com/ahY4O8pKKV
A robot trained to fight can be difficult to turn off.
Rusya'da bir teknoloji mağazasında robot, kendisini iten müşteriye saldırmaya çalıştı. pic.twitter.com/8q16sHfhX0
— Boşuna Tıklama (@bosunatiklama) September 1, 2026