mingos

relentless restlessness

[Patrick Collison @patrickc]

Great essay from @clarabcollier: https://t.co/H1LpUNid9O.

(You should subscribe to @asteriskmgzn, in addition to @WorksInProgMag of course. They have a very good print edition.)

[mingos @mingos3]

@mostik_ai ☀️☀️☀️

[mostik.ai @mostik_ai]

🔥🔥🔥

[👩‍💻 Paige Bailey @DynamicWebPaige]

🍎 Just learned that the team at @Orchard_Robots has been using @GoogleDeepMind's Gemini 3.7 Flash to power much of their work! They recently migrated from Gemini 3.1 Pro due to improved speed, lower costs, and improved performance with 3.7 Flash.

Orchard is combining rugged, tractor-mounted vision hardware (FruitScope) with modern farm management software to track precision data across billions of crops -- from grapes and apples (❤️!!) to almonds and citrus.

AgTech + frontier multimodal AI in the physical world is incredible to watch! @reservoirfarms

[Michelle Sun @michellelsun]

Inside Reservoir Farms @reservoirfarms w/ CEO Danny Bernstein @bernsteinres

America stopped applying for farm jobs.

"We are performing precision surgeries using robots, but we still hand harvest fruit."

Bernstein spent close to 20 years in Silicon Valley, including ten at Google, then built 40 acres of working farmland in Salinas where robotics startups test machines on real crops. Second site opened in Sonoma a month later.

"There were over 400,000 of these jobs that were posted in 2025, and only 182 domestic applicants."

"It's not AI for luxury, or AI for abundance. It's really AI for resiliency."

We cover:
› Why fresh produce still has no automated harvester
› The two-year payback every ag machine has to clear
› Who actually manufactures these machines, and where
› What breaks a robot in a field, and what everyone underrates
› Why the next site is going on a community college farm
› Add a crop, add TAM
› YC gives you a million dollars in tokens. Reservoir gives you a tractor.

TIMESTAMPS
00:49 Danny Bernstein, Reservoir, and a technical community for agriculture
00:49 Silicon Valley builds seven of everything, agriculture waits for one
03:15 Driscoll's, Taylor Farms, Dole, and a $4.8 billion county
05:29 Not AI for luxury, and the three gaps worth building into
05:29 Precision surgery by robot, fruit picked by hand
07:50 400,000 advertised farm jobs and 182 domestic applicants
10:11 What actually breaks a machine in a field
12:36 Sonoma, and the plan to cover the top ten specialty crops
12:36 Washington, tree fruit, and the Cosmic Crisp patent
14:58 Two years to return the cost of the machine
17:14 Service teams, dealers, and Andros Engineering
21:50 Merced College, and building on a teaching farm
24:12 YC gives you tokens, Reservoir gives you a tractor

[mingos @mingos3]

@michellelsun @reservoirfarms @bernsteinres Saved 👀

[Sohyeon Kim @ohmyksh]

We’re asking the research community to help us build a benchmark for research taste.

Scientific discovery starts with a fundamental step: which prior work is worth building on? We want to capture this undocumented layer through our collective knowledge.

Please sign up: https://t.co/T4vfwQnHUY ↓

[mingos @mingos3]

@sonyatweetybird @RichardSSutton Sutton argues as a philosopher, concerned with the problem not with the representation 🌺✍️

[Sonya Huang 🐥 @sonyatweetybird]

Today we release my favorite episode of Training Data yet: the great Rich Sutton.

@RichardSSutton wrote the textbook, wrote The Bitter Lesson (and many other on-point essays like "Self-Verification, The Key to AI"), and trained a mafia of talented students who went on to change the AI landscape forever including David Silver, inventor of built AlphaGo.

@kjaved_ was Rich's PhD student at Alberta and wrote The Big World Hypothesis. They just left academia to start @oaklab_ai

Their core argument:
(1) The Bitter Lesson: the world is massively more complex than any model of it, so anything trained on human-curated data has a ceiling
(2) Continual Learning: intelligence is continual by definition, and today's models stop learning the moment they ship.

The conversation covers:
— what The Bitter Lesson actually says, and what people get wrong
— why synthetic data is "just a big mistake," and the Big World Hypothesis behind it
— how LLMs are both a positive and a negative example of his own essay
— why no animal learns by supervised learning, and what squirrels can do that we can't
— the cure for catastrophic forgetting: per-weight step sizes and continual backprop
— why the biggest labs can't take a path where performance gets worse before it gets better
— a trillion parameters on 20 watts, and the Moore's Law math that makes it plausible
— why the endpoint isn't one mind but one design, running as many minds

It was both a fun generative idea- and debate-filled conversation, and a surprisingly human one too. Rich, thank you for beating cancer and changing the trajectory of AI. 💙

00:00 Introduction
02:10 An AI winter, a cancer diagnosis, and the move to Alberta
07:07 Writing "The Bitter Lesson," and what people get wrong
09:53 Are LLMs a positive or a negative example of it?
11:03 Synthetic data is "just a big mistake," and the Big World Hypothesis
18:01 AlphaGo, human priors, and why prior knowledge and learning should be friends
22:37 "Their weights never change": do LLM assistants actually learn?
26:09 Babies, squirrels, and why no animal learns by supervised learning
32:02 Rockets, imagination, and where paradigm shifts come from
36:42 The Alberta Plan and its 12 steps
38:53 Catastrophic forgetting and the cure
43:43 Oak's biggest ambition: a self-maintaining mind
47:56 Why the big labs are stuck in a local minimum
49:13 If everything goes right: LLMs, many minds, and hiring

The man who pioneered reinforcement learning thinks the rest of the field is weird, and lays it all out in today's episode. Together w/ @Alfred_Lin @sequoia

[mingos @mingos3]

@aimalysheva Great content, congrats 🎂

[Victor Escorcia @3scorciav]

🤣 Rebranding & ignoring some ppl are inherent to 𝐀𝐈.

“One reason for inventing the term [AI] was to escape association with cybernetics,” as McCarthy once bluntly explained. “I wished to avoid having either to accept Norbert Wiener as a guru or having to argue with him.”

[Ryan Dahl @rough__sea]

"The model is the computer" is my all-time favorite company mission statement.

I feel like Taalas's weight-printed-onto-chip is an idea that will shape LLM research in the future

[mingos @mingos3]

@michellelsun @gelsight Very insightful! 🫴

[mingos @mingos3]

"Tackling grand challenges requires genuine introspection and the unglamorous work of building understanding within and across communities to produce solutions with scalable impact"

[mingos @mingos3]

"Organizations must be redesigned around what AI makes possible: eliminating hierarchical bottlenecks so that workers can participate in a range of cross-cutting flash teams that allow operational data to continuously feed institutional learning"

[Blaise Agüera (@blaiseaguera.bsky.social) @blaiseaguera]

Our democracies, economies, and legal systems weren’t built to absorb AI at the rate it’s advancing.

In a new essay for @NoemaMag, our Reverse Alignment coalition map out the investments it will take to rethink the institutions, norms, skills, and governance frameworks needed for society to flourish with AI: https://t.co/uFhim9u2zN

[mingos @mingos3]

"doing this effectively requires harnessing these systems’ internal representations of the human cultural and relational dynamics that underlie research fields"

[Dezeen @dezeen]

Stadiums are becoming more distinctive with clubs and countries using them to attract attention. As part of our Future Stadium series, we look at why stadium design is becoming more icon-focused.

https://t.co/OBSSoUljov

[Valeriy M., PhD, MBA, CQF @predict_addict]

China has taken it to new level. Bejing AGI bar

[mingos @mingos3]

That's Minsky "The Society of Mind" approach
https://t.co/zPADHAGnEg

[Sasha Malysheva @aimalysheva]

the version of the future I'd actually defend: some constellation of specialized models communicating with each other, roughly the way parts of a brain do (one monolithic model that ends up being "the AI" is what I'd bet against)

and the hard part underneath it is the communication protocol itself: I think that's one of the biggest open problems we have right now

cutting the podcast into clips over the next couple of weeks, starting with this one 👇

[mingos @mingos3]

el cuello de botella actual es la falta de mecanismos para convertir datos externos no estructurados en supervisión robótica fundamentada

el problema no es solo "más datos", sino su interfaz: hacer que esa experiencia física sea legible y aprendible por los robots

[Michelle Sun @michellelsun]

"Robots Need More than VLA and World Models" is a necessary read for anyone mapping the physical AI stack.

The paper takes a hard look at the core bottleneck in robotics.

It’s the 𝗴𝗿𝗼𝘂𝗻𝗱𝗶𝗻𝗴 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸: our inability to convert the world's abundant unstructured physical data (like internet videos or human motion) into grounded robot supervision.

To train robust policies, we need data rich in action labels, task semantics, and reward structures.

To bridge this gap, the authors propose moving beyond end-to-end scaling to build four critical pillars:
𝟭. 𝗗𝗮𝘁𝗮 𝗘𝗻𝗴𝗶𝗻𝗲: Autolabel unstructured behaviors (like passive video) into high-quality training data at scale.
𝟮. 𝗘𝗺𝗯𝗼𝗱𝗶𝗺𝗲𝗻𝘁 𝗜𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲𝘀: Retarget task-relevant human motions or cross-embodiment data into robot-specific actions.
𝟯. 𝗪𝗼𝗿𝗹𝗱-𝗠𝗼𝗱𝗲𝗹: Provide physics-grounded 3D reasoning to predict the physical consequences of actions before they happen.
𝟰. 𝗥𝗲𝘄𝗮𝗿𝗱 𝗜𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲𝘀: Infer task progress and success from video and language, enabling the self-improving deployment loops needed to handle edge cases.

When we transition from relying solely on scarce, robot-native datasets to leveraging world-scale physical supervision, systems can truly begin to learn from the physical world.

If you’re building hardware, VLAs, or the data pipelines powering robotics foundation models, how are you thinking about mapping out these interfaces?

I broke down the compounding loop in the graphic below. Paper link below: