Jaana Dogan ヤナ ドガン pfp

Jaana Dogan ヤナ ドガン

@comfyre

354 Following
333 Followers
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
Advice @simonw gave on the podcast episode about AI tools: Every developer should try out a try out a small, local LLM model! Download and run from eg HuggingFace. His point is you understand limitations of *all* LLMs much faster with small models. Like hallucination! (I agree)
0 reply
1 recast
4 reactions
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
If you want to have a solid and coherent API, you really need a few opinionated API designers. Anything else is often counterproductive.
0 reply
0 recast
3 reactions
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
🧬 This computational synthetic biology paper highlights something under-appreciated about JAX: JAX code is similar to numpy, with outsized performance and scaling benefits - which means that researchers can run experiments that would otherwise never have been attempted. 🥹❤️
0 reply
2 recasts
3 reactions
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
Much to learn from this election but I’m not sure if people will ever agree what the learnings were.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
Me: What am I ordering? Glenmorangie 18 year old? Him: No, no. Get the largest bottle of Jameson.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
This culture is partially gone for some number of valid reasons, but I fear we went too far now that some engineers feel ashamed publishing anything if it doesn’t look complete on day 1. So much opportunity is lost because perfect is always the enemy of good.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
It’s possible that Google’s most major problem is self inflicted gate keeping of its own tech. Earlier Google was a inspiring infra builder because it was easier to build publicly. Network effect created major communities around several of our foundational tech initiatives.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
Ballot is ready to return. The future of our city is also on the ballot. The margins in local elections are tiny and your vote matters.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
I genuinely think that human progress may exponentially grow to the point that we may end up surviving long enough to see humanity finding a way to defeat death. Then, it will be so obvious to me that we are in a simulation, yet it’ll be too disturbing to accept that.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
Mail: Cloud budget exceeded. Me: 😱 Am I bankrupt? Mail: Your budget was $1.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
They’ve been cooking for a while, this was the keyboard setup when I was there back in 2015.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
It makes no sense to build AI infra in a closed fashion when hundreds of startups are building something in the open.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
Nuclear power having a real moment again has been one of the best outcomes of the AI ride so far. Some of us feel like we can forgive the personal pain and burnout for this particular outcome.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
We're hiring! ;)
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
Multi-region, scale-to-zero, on-demand, pay-per-use GPU AI inference with @ollama? gcloud beta run deploy --image ollama/ollama \ --port 11434 \ --gpu 1 \ --region us-central1,asia-southeast1
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
OH at a meeting: Had to cancel a meeting with my boss because he need to take a moment to celebrate his Nobel.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
I’m glad I didn’t have the intellectual capacity to study physics, because computer science became physics in the end. If you wait enough, some problems disappear.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
The downside of building your own in-house infrastructure at the scale of Google is when the original authors leave/retire, people cannot connect the dots back to the original ideas and tradeoffs. It causes wrong assumptions and decisions in the long tail.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
This field is the hardest I’ve ever seen. I’m super glad that we work with some of the most solid people on earth. I feel extremely lucky and it’s been a while since I’ve felt this way.
0 reply
0 recast
0 reaction
Jaana Dogan ヤナ ドガン pfp
Jaana Dogan ヤナ ドガン
@comfyre
"Why do people in AI move around so much?" People follow the leader, hardware, product... Like all of these dynamics change every other 3-6 months.
0 reply
0 recast
0 reaction