The equivalent of a rough night out in San Francisco is realizing your company has burned through its annual token budget last night.
Stop the "morning-after-bill". Get your own model, build your own intelligence. Reach out and more to come soon. https://t.co/JVwJvUWOqx
Anthropic's Claude Design launch blindsided its partners, Figma and Canva.
“Essentially, Anthropic had kind of told these partners like Figma and Canva that the Claude Design product was going to be fairly basic.”
“As it got closer to the launch, these partners found out that actually this new product would have some of those advanced features [Anthropic said wouldn’t be in there].” — @steph_palazzolo, AI reporter
After coding is solved, the next frontier is computer use. Today, we are launching Use Computer, the infra for evaluating and training models to use all kinds of computers 👇
Please RSVP with the link below if you are in Amsterdam. Discussing methods on how to build AI that understand design and human preferences.
Also more to come later this month... 👀
#bryel#ai#posttraining#taste
Really excited to be hosting a AI research dinner with Bryel Labs. At @framer we're exploring how to build models with taste, and we'd love to hear your thoughts over dinner.
Apply here: https://t.co/4P243SE0lO
Your data is your edge, but only if your AI is built on it. Rent a generic model and so can your competitor. The companies with an edge are deploying custom models that they own and improve over time.
Our co-founder @rhythmrg recently stopped by @southpkcommons to share how companies are owning their intelligence with Applied Compute.
I truly reject the premise that this will be sustainable however. This is a transient period in which there is an information asymmetry between the top AI labs and the companies on the ground.
FDE job openings are up 10x YoY!
I sat down with @colintjarvis, @calvinleenyc, @jrshoch, and @howard_zuo to discuss why top AI companies are bringing their best engineers on-site with customers.
One reason: coding agent empower engineers to focus less on integrations/grunt work and more on solving end-to-end customer problems, where on-the-ground context is critical.
Great to see that we are finally paying attention to train models to understand humans/develop benchmarks around it, rather than the other way around.
Great work @karinanguyen & team
In post-training, we've learned that once a behavior is measurable, you can train AI to excel at it.
EQ is one of the hardest things to verify. AttuneBench makes it measurable through observable signals: whether a model notices distress, tracks shifting preferences, adapts to context, and responds in a way people experience as helpful.
UC Berkeley Post-Training Demo Day at @agihouse_org on 4/27 from 5-7:30pm.
Come through and see what agents our students have built. 👀
s/o to my co-instructor and sponsor @karinanguyen
https://t.co/zUNCjrX7bH
Gemma 4 looks at a parking lot. Decides what to ask. Calls SAM 3.1.
"Segment all vehicles." 64 found.
"Now just the white ones." 23 found.
One model reasoning and orchestrating. One model executing.
Both running locally on a MacBook. MLX. No cloud. No API.