💯
I get that everyone is different and it may take time to ramp up properly so that it's not just token maxxing for the sake of token maxxing.
But $4k - 10k on tokens per month is entirely reasonable and preferable than starving a talented engineer. Way more for the 100x'ers.
A $500k engineer costs $2,000 per working day.
If $200/day of LLM tokens makes them 20% more productive (which is absolutely true today), that’s one of the easiest ROI in the company.
The expensive thing isn’t tokens. It’s underutilized engineers.
Another one for the attackers advantage. Defenders often move much, much slower. Bureaucracy / process, plus risk in breaking unintended things.
My main viewpoint is that attacking is verifiable and defending never will be. So AI attackers always have an advantage.
1/ You can now try Cua-S1-4B-0.2 in your browser. Thanks to @multimodalart at @huggingface for building the Space!
Pick an example and see how the model scores the candidate actions.
Australia has been hacked.
'And today, I spoke with the CEO of OpenAI, Sam Altman, to express Australia's extreme concern about this incident. And I also expressed my disappointment that it took the company way too long to inform the government what had occurred, and the nature of the way that that notification occurred as well was unacceptable.'
brb joining oai to fix their fucking ios apps
can’t even have a chatgpt thread make a bug report because that crashes the app
agi is not here if they’re still using electron
Alert: Apple just dropped a new model on Hugging Face.
It's a Qwen3.5-9B finetune that turns long documents into small page images to save tokens, then pulls up the full text of only the pages relevant to your question 💡
1/ Today we're introducing Cua-S1-4B-0.2, the first multimodal decision model trained with RLOO on live computer-use tasks, using task-completion rewards.
Text and multimodal adapters are available under Apache-2.0:
We gave ChatGPT, Claude, and Grok control of a real Toyota Corolla 🚗, steering/gas/brakes, no human driving (just a foot over the brake).
Only one model was able to complete our entire driving course. Introducing DrivingBench. 🔥⌛️🏁
Building a custom ebike stair ramp for my basement staircase access with astra. Did a bunch of lidar scans and manual measurements and pictures. Will figure out material sourcing and give it a try.
No idea if this will work but fuck it, lets ball. My ebike is way too too heavy.
We've got models that _can_ unlock whole new worlds and impressive workflows from their intelligence. I think there's still a lot more we can push these models to do that most of the world has not unlocked yet.
However, what have we not even imagined yet if speed and cost significantly improve?
That's where we're going, and it's equally as exciting. Jev was but a small taste of that. I'm having a great time with open flash models now, I'm exciting for this to improve with Opus 5.5, Sonnet 5.5, and these Luna models.
Googlebook is officially here! I’ve been so excited to share the details with the world.
Today, many of us rely heavily on laptops to get work done, but we think there is an opportunity to rethink the category to address the needs of people today. So we brought the best of ChromeOS and Android to create a new platform for laptops.
Our initial focus with Googlebook is to deliver an amazing laptop that feels awesome for Android phone users.
Here’s what to know about Googlebook 🧵👇