@pavluxa Profile picture

Pavel Dournov

@pavluxa

Software Engineer, AI systems. Google, Microsoft, Azure, Cloud. PNW resident, coffee and running enthusiast. Opinions are my own.

Joined July 2008
Similar User
South Park Commons photo

@southpkcommons

Walter Scheirer photo

@wjscheirer

R. Burke Squires photo

@burkesquires

Jeremy Lewi photo

@jeremylewi

Aiforia photo

@aiforia_tech

Mohamad Amin Mohamadi photo

@QuelMohamadAmin

π12Qα photo

@firqaaaa

Aswin Raghavan photo

@aswin_raghavan_

Laksana Utama photo

@Aiolia21

matt brown photo

@mnbbrown

Hai Nguyen photo

@hainm_comp

Gunleik Groven photo

@gunleik

J photo

@j__io

ohnabe photo

@ohnabe

Pavel Dournov Reposted
Tweet Image 1

Pavel Dournov Reposted

Precisely


Is there any reason to work on anything else but AI? AI will do the rest.

We're starting to roll out API support for Gemini 1.5 Pro for developers. We're excited to see what you build with the 1M token context window! We'll be onboarding people to the API slowly at first, and then we'll ramp it up. In the meantime, developers can try out Gemini 1.5…



Pavel Dournov Reposted

Less than 24 hours ago, Anthropic announced Claude 3. And some people are calling it better than GPT-4. Some are "feeling" Claude 3 (Opus) has reached AGI level. 8 incredible examples:

Tweet Image 1

Pavel Dournov Reposted

The Gemma-7B model exceeds the performance of the widely used Llama-2 7B and 13B models on 8 of 8 benchmarks covering general language understanding, reasoning, math, and coding.

Tweet Image 1

Pavel Dournov Reposted

From a DSPy standpoint, very happy to see a 2B size—and a 7B size that works that well. Will enable a lot of cool stuff.

Introducing Gemma - a family of lightweight, state-of-the-art open models for their class, built from the same research & technology used to create the Gemini models. Blog post: blog.google/technology/dev… Tech report: goo.gle/GemmaReport This thread explores some of the…

Tweet Image 1


indeed. the simplicity makes you think why wasn't it done before...

Introducing Gemma - a family of lightweight, state-of-the-art open models for their class, built from the same research & technology used to create the Gemini models. Blog post: blog.google/technology/dev… Tech report: goo.gle/GemmaReport This thread explores some of the…

Tweet Image 1


Can’t agree more. Immediate massive mindset shift.

Introducing Gemma - a family of lightweight, state-of-the-art open models for their class, built from the same research & technology used to create the Gemini models. Blog post: blog.google/technology/dev… Tech report: goo.gle/GemmaReport This thread explores some of the…

Tweet Image 1


Pavel Dournov Reposted

Leslie Lamport just published a draft of his new book. 🔥

Tweet Image 1

Pavel Dournov Reposted

. @zacharynado pointed out this stat we'd put in the white paper that in retrospect deserves calling attention to. During our Gemini Ultra training run, we had a goodput measure of 97% (goodput: the time spent computing useful new steps over the elapsed time of the training…

this program just proved yet again that Google has the best systems infra teams in the world, hands down, getting us an insane goodput of 97% for the Ultra training run

Tweet Image 1


Pavel Dournov Reposted

I’m very excited to share our work on Gemini today! Gemini is a family of multimodal models that demonstrate really strong capabilities across the image, audio, video, and text domains. Our most-capable model, Gemini Ultra, advances the state of the art in 30 of 32 benchmarks,…

Tweet Image 1
Tweet Image 2

Pavel Dournov Reposted

A nice post about the various techniques used to get a training job to scale to more than 50,000 TPU v5e chips on @googlecloud cloud.google.com/blog/products/…


Congratulations to Pavel Dournov on completing 2023 Snohomish River Run. 136th, 1:50:25.6. See more… results.raceroster.com/v2/en-US/resul… via @RaceRoster


Pavel Dournov Reposted

Does everyone need the ability to provision tens of thousands of TPUs to train AI models with trillions of tokens? Of course not. Few do. But the concept here of "Multislice" may prove useful to many, and the tech details in this post are terrific. cloud.google.com/blog/products/…

Tweet Image 1

One of the most advanced, scalable, secure and performant AI infrastructure coming to GCP! Expanding our AI-optimized infrastructure portfolio: Introducing Cloud TPU v5e and announcing A3 GA @googlecloud cloud.google.com/blog/products/…


I have had my Mac M1 for years and just now turned on music on it. The sound quality is just amazing - wide range, deep base, immersing stereo - great piece of engineering in these tiny embedded speakers.


Hobby coding nowadays - first create prompts for chatGPT and cloude, define the problem clearly. Generate code, inspect, run. Iterate. The version control is more important on the prompts.


You learn so much from carefully reading a 30+ comment thread in a bug report, looking up every acronym or new term.


Loading...

Something went wrong.


Something went wrong.