MoStark

wanna talk to your drone like I do? 👇

[MoStark @MoStark_ai]

Just uploaded ChatGPT to my drone

[MoStark @MoStark_ai]

just reached my weekly limit for Wispr too

[MoStark @MoStark_ai]

@fraserpricee @mitsuhiko perhaps they're afraid of getting gamed

[MoStark @MoStark_ai]

@lydiahallie that's really great, feels like someone finally listened :)

[MoStark @MoStark_ai]

The AI benchmark that marks the end of the world?

July 24, 2026,

Andon Labs, in collaboration with Anthropic, publishes Drone-Bench.

The idea is to evaluate the extent to which current models are capable of making a drone autonomous in order to allow it to track a person, reconstruct in 3D the environment in which it progresses, etc.

At the time of publication, none of the Claude Opus and Fable 5 models managed to pass the test ❌

September 10, 2026,

Andon Labs, this time independently, replays the benchmark on GPT-6 Astra and announces that this model has passed all the tests.

✅ or ⚠️?

September 12, 2026,

Dario Amodei publishes We Must Pace the Frontier ⚠️⚠️⚠️

He warns us of the risk of military domination via AI drones. Fear is legitimate.

A little trip back in time...

Since June 10, 2026,

A few months before all this, I post videos on YouTube / Instagram / TikTok of my drone co-pilot voice agent project (thank you ElevenLabs), under @ mostark_ai.

My videos have gone viral, now accumulating more than 4 million views.

And I could see that the sentiment has shifted in the last few days from the excitement of finally seeing something other than AI slop on the networks to constant fear.

And personally I think it's a shame...

I received a lot of messages from people who shared with me their pain point with drone apps and saw in my initiative hope for a better solution.

In particular, I was contacted by:
- firefighters in the Landes and Fontainebleau regions in France. Both had the same need: find a solution to help them detect forest fires from the air
- a person who became quadriplegic and whose dream is to fly a drone
- an American startup that offers a solution to help municipalities find missing people

So what do we choose: fear or hope?

I partly share Dario's view. Obviously there are issues that I am very far from (Paris to Taipei is 9,804.4 km hum hum), but I see the value of mastering these AI systems with the right safeguards.

Last anecdote: initially, I had directly connected the AI to the DJI SDK function that makes the drone fly and as a result, I crashed it against the ceiling. I clearly learned harness engineering the hard way 😅

Pace, yes, to have time to secure, but prohibiting progress in the name of a few excesses? I don't agree. And given what happened in the ballot boxes for the EU's AI Act this summer, I'm not the only one of that opinion.

Peace.

MoStark.

[MoStark @MoStark_ai]

@gpumaxxer damn

[MoStark @MoStark_ai]

@thsottiaux dude

[DAIR.AI @dair_ai]

Impressive paper showing the impact of a good harness.

Improves Qwen3-8B from 41.2% to 91.8% on long-horizon robot tasks without any change to the model.

The gain comes from the harness.

GAVEL keeps an explicit graph world model holding object relations, action preconditions and effects, and probabilistic beliefs about where unobserved objects are.

Before the robot executes an LLM-generated action, the graph predicts what that action would do.

Violations get caught, and the ones whose fix follows directly from the world model get repaired without calling the model again.

Only errors that need semantic reasoning go back to the LLM.

On BEHAVIOR-1K across 500 multi-task instructions, success rises from 19.9% to 92.6%. Reasoning over the distribution of possible object locations also reorders the remaining subtasks and cuts travel distance about 5.4%.

Why does it matter?

Many long-horizon agent failures are state-tracking failures rather than reasoning failures, and a symbolic model sitting outside the LLM catches them cheaply.

Paper: https://t.co/WcQ7ygGkMF

[Yann LeCun @ylecun]

I said "auto-regressive LLMs, in and of themselves, will not lead human-level AI"
That statement is still totally true.

First, the reasoning abilities of current AI systems are based non-auto-regressive search (which is what I have always advocated for). But AFAICT, they do it in token space, which is limited and inefficient. I have claimed that human-like reasoning must be a search in continuous representation space. It looks like the industry is moving towards that.

Second, the self-improvement methods, as currently practiced, only work for domains where the quality of outputs can be scored without human intervention, such as mathematics, code, and scenarios that can be simulated accurately. Not anything else. Humans and animals learn new skills way more efficiently than current RL methods.

Third, the multimodal capabilities of current AI assistants generally use separately-trained encoders (that are not LLMs). This is also what I've been advocating. Except that I think the best way to do this is with JEPA trained with self-supervised learning. The research community is clearly moving towards that (3000 papers on JEPA in just 4 years).

Fourth, if LLMs were a path to human-level AI, we would have domestic robots and Level-4 or Level-5 self-driving cars for consumers by now. And we don't. We certainly don't have cars that can learn to drive in 20 hours or practice like any teenager. We're still missing something pretty huge to claim human-level intelligence (let alone superhuman).

Sure, we now have computer systems that are impressive, very useful, and whose performance is superhuman in an increasing number of domains (coding being one of them).
But that's true of the entire history of progress in computer technology.

Lastly, there is a basic confusion about what intelligence actually is.
It is not the mere accumulation and regurgitation of existing declarative knowledge (which is essentially what LLMs do).
As Jean Piaget famously said, "intelligence is not what you know, it is what you do when you don't know."
It is your ability to solve new problem without any prior training, to act in previously-unknown scenarios, and to adapt very quickly to new situations with minimal training.

We're still far from that.

[MoStark @MoStark_ai]

exhausted my codex weekly limit in 6h btw

[MoStark @MoStark_ai]

@TaylorBereiter @Anthropic @notjazii @threejs this is sooo amazing, been playing for 10min it's very clean

[MoStark @MoStark_ai]

@Tech_girl @Im_IrushiK literally shaking rn

[MoStark @MoStark_ai]

/btw on codex >> /btw on claude code

[MoStark @MoStark_ai]

@hegargarcia tell it to change maxOutputTokens from ∞ to 3000 or so

[MoStark @MoStark_ai]

@AestherML @thsottiaux @hiarun02 I mean, I'm no expert but that would cost++ no?

[MoStark @MoStark_ai]

@ncfrontiersman @thsottiaux @hiarun02 Complaining about what? I just crushed a week’s tokens in 6 hours 😂

[MoStark @MoStark_ai]

@realDonaldTrump Mr president, what about America Intelligence?

[MoStark @MoStark_ai]

@thsottiaux @romainhuet @sama any reason why I can't choose the 20x offer? I'm in France btw

[MoStark @MoStark_ai]

@thsottiaux @romainhuet @sama I see signs of reset but I can't prove it

[MoStark @MoStark_ai]

@CompleteSkeptic, is Jev running on a quantum computer? don't lie to us