* Co-Founder of Nora * Co-Founder of @wendylabsinc *Co-Founder of https://xitter.cf/t.co/Prt2fRyKQj @dittolive

San Francisco, CA
Joined January 2015
Maximilian Alexander retweeted
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine. • Best-in-class fast local coding • Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark • Built from scratch in under 4 months Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
618
883
618
8,343
3,358,921
Maximilian Alexander retweeted
After being at the physical AI summit in SF, I am convinced that the bitter lesson that coding harness solves the world applies absolutely to every industry every single one
15
6
2
225
28,557
Maximilian Alexander retweeted
New on our Science Blog: designing protein binders against scorpion neurotoxins. Scorpions sting about 1.2m people every year, and antivenoms are often costly or in short supply. Claude has made an early step toward finding new treatments. anthropic.com/research/claud…
496
616
369
7,329
1,787,427
A slide from my first startup; when this was considered an obscene seed.
1
11
333
Maximilian Alexander retweeted
i'm teaming up with @vivmarquez and @liquidai for an on-device AI hackathon!! take a @Qualcomm_Dev IQ9 EVK and any of the LiquidAI models and build something that works in Airplane Mode! i'm pumped!! it's going to be awesome and i can't wait to see what folks build. more details here: luma.com/0nsdnkoa
3
2
1
11
698
Maximilian Alexander retweeted
I feel Agentic AI and Robotics will eventually converge. Both are about reasoning and acting, just in different environments. Acting in the physical world is much more challenging, involving embodiment, perception, control, safety, and physical interactions.
9
8
1
100
4,001
Maximilian Alexander retweeted
May not be much, but generalizes pretty well and is fully autonomous and doesn’t use the evil VLAs
1
3
55
2,735
Maximilian Alexander retweeted
This doesn’t seem to be a very robust strategy in the era of AIs that can use your computer on your behalf.
notice.. > Delta is blocking AI agents from booking flights > JPMorgan Chase and Charles Schwab block agents from accessing accounts > Yelp requires paid data licensing for agents > Amazon has blocked Grok bot and other shopping agents. the internet is starting to block AI agents that were supposed to do everything for us
158
25
8
781
70,599
Maximilian Alexander retweeted
Developers are going to be a small part of total token usage over time, as the internal stats on usage at OpenAI and Anthropic both show. And for long-running high value tasks in areas where verification is harder, the benefits for higher "intelligence" models are quite large.
OpenAI, Anthropic and other leading AI labs face a growing challenge: Better models aren't necessarily attracting more usage. A 10-point intelligence advantage now earns just 20% more developer traffic, down from nearly 69% in late 2025. Full story: thein.fo/4hGJ9jY
162
39
10
660
45,382
Maximilian Alexander retweeted
This is what you can do with: - Good engineers (@SamuelSBlackman, @OliverWTaylor ) - Good tools (@wendylabsinc) - Good LLMs (@OpenAI) – at Best, Nederland
These past 2 weeks have made RL a lot easier. I made a really nice mini golf sim in >2 days. A good LLM makes setting up IK demos → behavior cloning surprisingly straightforward. Action chunking has made my motions much smoother. Just ensure that you are adjusting K values and chunks according to your real robots Hz (usually camera or policy or joints have a maximum and will be the bottleneck) SGS helps decide what to practice next instead of sampling everything equally. Surprisingly makes training very reliable. PPO still gives me the best results. I still have faith that someone will make SAC with better adjustments but that doesnt matter. Still unsure where self play is the go to method for most things but I usually use this for all things RL. Also, people really dont need GPUs that are too expensive. With motion priors, training should almost always take under 24 hours to see solid results.
2
3
409
Maximilian Alexander retweeted
The games we've long wanted, which didn't exist because they'd cost several times more to make than they'd earn back, will soon become possible. As will the games developers wanted to build, but couldn't with the resources available. It'll be a new golden age of gaming.
2,161
1,306
917
17,635
5,715,641
Every month, I open up Siri and I celebrate Apple for not going into AI. Also extremely bullish that Ternus will announce updates to Muse on Apple silicon than Siri. I’ve never seen this company hate their product so much and get lauded.
Microsoft and Meta are pulling back on AI spending. Anthropic investors are now talking $1.5T for the IPO - not $2T. Apple is being called an AI winner for never entering the LLM race. Something is happening.
1
3
295
Maximilian Alexander retweeted
LMAO, #Blender is finally free and open source!
525
1,201
318
23,025
861,792
Maximilian Alexander retweeted
twitter is the bloomberg of ai research
135
294
34
4,761
103,088
Maximilian Alexander retweeted
Over the past month, we completely switched to frontier LLMs to control a robot in a customer deployment. By doing this, we have a classic System 1/System 2 architecture, but on steroids. For the high-level reasoning layer, we use an agent based on Opus and Astra, which are many generations more capable than any VLA. For the fast execution layer, we use computer vision code written by the model, which is fast, editable, and reliable. As a bonus, we got in-context learning, which is just a basic feature of LLMs, and continual learning, since whenever a new situation arises, the large model steps in and adapts the pipeline. And it just feels better to be on the right side of the Bitter Lesson, benefiting from the billions being invested in frontier models around the world.
51
70
30
1,075
92,702
Maximilian Alexander retweeted
Following three years of development, FluidNinja LIVE-2 released - free to download at the community server #FluidNinja #UE5 #realtimeVFX #VFX #gamedev
89
432
40
4,283
501,159
Maximilian Alexander retweeted
if i leave the apple ecosystem it's going to be because of Siri
386
208
73
11,654
336,788
This is not common sense. Everyone knows you cannot tolerate a team that ships stuff slower and worse than if the ceo prompted it.
Everything about how we build and scale businesses seems to be changing at an unprecedented pace. Long-held assumptions are being challenged, and many of the old playbooks are being thrown out. But the more things change, the more they stay the same. Technology changes, business models evolve, but the fundamentals of great leadership don't. I shared this advice with our company's senior leaders today. Three of the worst things a leader can do are refuse to hear bad news, avoid conflict, and try to please everyone. Why is this the case? If you don’t want to hear bad news, you can’t solve problems. And solving problems is one of a leader’s most important responsibilities. Ignoring problems doesn’t make them go away. Avoiding conflict only creates more conflict. Unresolved issues fester, frustrations build, and relationships deteriorate. Trying to please everyone ultimately pleases no one. People quickly recognize when you say different things to different people, creating a false sense of alignment that erodes trust. The litmus test of any team, especially a leadership team, is its ability to have hard conversations about real problems and come together to resolve them. That requires trust, candor, humility, and above all, courage. The courage to hear uncomfortable truths, confront difficult issues, put the team’s success ahead of individual egos, and tell people what they need to hear, not just what they want to hear. This sounds like common sense, but as we know, common sense isn't very common because it requires emotional fortitude. Looking back on my career, the toughest leadership decisions never tested my intelligence. They tested my courage.
1
3
224
Maximilian Alexander retweeted
This paper shows why I am skeptical of the idea that frontier models can be easily replaced by cheaper ones (and why you should use the best models as orchestrators): Small improvements in accuracy & judgement compound over long-horizon tasks, creating huge gains in outcomes.
197
112
28
1,224
67,069
2027 will be a lot less about general coding -> coding will focus exclusively to cyber security. But the biggest winners will just be Codex and Claude Code. Using a wrapper service won't be helpful After cyber all focus will go to scientific discoveries (which is largely robotics and physical AI)
1
1
133