abu

I have a single resolution this year. Every year, I experience this palpable feeling of a fresh slate. All the baggage from the past year, dropped. My goal is to feel that every morning this year. To remind myself that every second alive is an opportunity to change direction.

some of the rejected (but still beautiful) alternative choices:

@fanjiewang you guys rock

been working with these guys since the early days of Zen always delivered 100+ TPS and sub-second TTFT across various models we’ve collab'ed on

Please, just call me dawg. Big dawg was my father

@philipkiely so frickin cool, kudos philip!

Inference Engineering launches today. baseten.com/inference-engi…

b10’s inference infrastructure powers the most sophisticated app companies serving specialized models at scale. there are dozens of these companies today and there will be thousands in the future. incredible momentum + world-class talent density. and, it’s still so early…

IMO it should be considered quite rude in most contexts to post or send someone a wall of 100% AI-generated text. “Here, read this thing I didn’t care enough about to express myself.”

@ArtificialAnlys try it here baseten.co/library/glm-4-…

happy holidays! we just dropped the fastest GLM 4.7: 400+ TPS as benchmarked by @ArtificialAnlys

wander and wonder

they're really so good. @saltyph and i spend about 10 mins of every 1:1 raving about them and imagining all the magical experiences we are going to build together.

Today we’re welcoming the @parsedlabs team to Baseten! With their RL and post-training expertise, Baseten is enabling companies to own their intelligence by unifying training and inference. Read more about it from the Parsed founders ⤵️ baseten.co/blog/parsed-ba…

@parsedlabs has been acquired by @baseten. Big Token wants you to believe the future is a monoculture: one model to rule everything, one bill to pay forever. Rent the demigod, trust that next month's update will finally solve your problem, and pray that GPT-(n+1) happens to have your exact behaviour covered in its pre-training data and RL environments. It's a convenient story for their bottom lines. We took the other side of that bet. A year ago, @mudithj , @maxkirkby and I were three Australians with a thesis that felt unfashionable: for the real, task-specific work, a specialised model will win. There are learning signals from the real world that you can't distil into a prompt. We thought the frontier was flattening and that the next wave of value would come from depth, not scale. Models that actually understand (and remember!) the job they're hired to do, trained on real feedback, owned by the people who use them. Democratised Token, rather than Big Token. We were right about the thesis. We were wrong about how hard inference is. Every customer deployment taught us the same lesson: training a model that outperforms GPT-n is only half the problem. The other half is running it fast, reliably, and at scale. Baseten was the only one that could actually do it. Eventually the question became obvious: why are we pretending this is two companies? To @tuhinone, Phil, @amiruci, Pankaj, and the entire Baseten team: you've quietly become one of the most important companies in AI, and we're thrilled to help make that louder. To the Parsed team: you are the most technically brilliant and genuinely good group of people I've worked with. Every late night, every commute to work (about 15 ft from our rooms to our desks), every argument about loss functions and Apple Monitors at 2am, it all mattered. To Julia and Ash from LocalGlobe: thank you for believing in our contrarian vision when it was challenged the most. To Max, this is our latest adventure, and none of this would be possible without the technical brilliance and hardened experience you've developed over the time that I've known you. And finally to Mudith: you are the best CEO and most authentic leader I could have picked to found this company with. I'm lucky to have built this with you both. We set out to help models touch grass. Now we get to do it with more compute, a bigger team, and the best inference stack in the world. Time to land some clean shots on the crowned giants of San Francisco.

@ZachyQ stud

Old film #3

@runwayml open source this

Today we're sharing our first research work exploring diffusion for language models: Autoregressive-to-Diffusion Vision Language Models We develop a state-of-the-art diffusion vision language model, Autoregressive-to-Diffusion (A2D), by adapting an existing autoregressive vision language model for parallel diffusion decoding. Our approach makes it easy to unlock the speed-quality trade-off of diffusion language models without training from scratch, by leveraging existing pre-trained autoregressive models.

@vim_dzl u not

Biggest fear about TLPM is that I'm doing both jobs poorly

We're changing our name. Captions is now Mirage. Today's most popular medium is short-form video. It's the fastest-growing format, it’s what people want, and it’s what brands need to grow. But unlike other forms of AI content like text or image, which can be generated and ready-to-use with a single prompt, generated video still requires a lot of manual work and editing to be ready to share. This is where the real opportunity lies. The way we see it, the real race for AI video hasn’t even begun. The world is waiting for a foundation model for video that can take an idea and produce a fully usable, cohesive output from scratch. So that’s what we’re building. We’re rebranding to Mirage to reflect our expanded vision for video, starting with the short-form format, through frontier AI research and models. Check it out at mirage dot app

@holynski_ holy

Something fun we discovered: you can use #Genie3 to step into and explore your favorite paintings. Here's a short visit to Edward Hopper's "Nighthawks".

@DennisHXu @tuhinone @OpenAI 200+!

@tuhinone @OpenAI congrats man, how many tokens per second are you guys running it at?

We're very excited to be an @OpenAI launch partner for GPT OSS. Today's a big day for open models, and we have day 0 support for GPT OSS 120b via our Model APIs: baseten.co/library/gpt-os… We'll be rolling out more performance optimizations and benchmarks over the coming hours and days, so stay tuned -- and congrats to the OpenAI team on the launch!