Dmitry (Dima) Lepikhin

Gemini Pretraining co-lead

[Dmitry (Dima) Lepikhin @lepikhin]

We have amazing cadence of pushing frontier forward!
*hiring in Performace (team is industry SOTA by a big margin)

[Arena.ai @arena]

Gemini 2.5 Pro #1 across ALL categories, tied #1 with Grok-3/GPT-4.5 for Hard Prompts and Coding, and edged out across all others to take the lead 🏇🏆

[Dmitry (Dima) Lepikhin @lepikhin]

@dustinvtran Length penalty is all you need paper confirmed

[Dmitry (Dima) Lepikhin @lepikhin]

@giffmana Hmm i can guess OP from this post

[Dmitry (Dima) Lepikhin @lepikhin]

@gabriberton ESL ppl from the Eastern Europe often say Gemin-ee

[Dmitry (Dima) Lepikhin @lepikhin]

@anmol01gulati gl hf

[Dmitry (Dima) Lepikhin @lepikhin]

@elonmusk Elon asking grok which model to train next confirmed

[Dmitry (Dima) Lepikhin @lepikhin]

@giffmana Is Gemini benchmark overfit model?

[Dmitry (Dima) Lepikhin @lepikhin]

@elonmusk 10T total? Activated?

[Dmitry (Dima) Lepikhin @lepikhin]

@GothamChess You still did not post on Fabi : Giri! Its was 1h ago!

[Dmitry (Dima) Lepikhin @lepikhin]

@VCBrags @Vullety One early investor in uber did

[Dmitry (Dima) Lepikhin @lepikhin]

@bgurley @leixing77 @grok this is not lidar tho, digital lidar is protein leather

[Dmitry (Dima) Lepikhin @lepikhin]

@VCBrags @hamptonism Milano is hot af, but great coffee, Paris coffee is a joke

[Dmitry (Dima) Lepikhin @lepikhin]

@giffmana @lossfunk Whats salami publish?

[Moritz Kremb @moritzkremb]

There's finally a proper benchmark for @openclaw model performance.

I just found that @kilocode built an open source benchmark that tests models across 23 real world openclaw tasks like scheduling meetings, writing code, triaging email etc

gpt-5.3-codex is sitting at number one. tbh that matches my experience.

gemini 3 flash in second place. didn't expect that.

curious to see where gpt-5.4 will land on this.

[Dmitry (Dima) Lepikhin @lepikhin]

@vonderleyen 🤌😘

[Dmitry (Dima) Lepikhin @lepikhin]

@tymrtn It's quite obvious that distillation is there for any non-frontier model. Web has plenty of traces from frontiers.

[Dmitry (Dima) Lepikhin @lepikhin]

@AnthropicAI This is unsurprising, thankfully it's not the tactic that produces new frontier.

[Dmitry (Dima) Lepikhin @lepikhin]

@auren Include pot

[Dmitry (Dima) Lepikhin @lepikhin]

@agihippo Well, the issue is, I'd like to do a coffee place like in Italy, hole in the wall, 1 euro caffe normale at the bar stand, maybe ocasional cappuccino, but there is no place for it in Bay Area.

[Anselm Levskaya @anselmlevskaya]

AI folks radically overestimate how much LLMs help for practical bio lab work and so get weirdly fixated on biorisk scifi scenarios. Lab work is gated by a researcher's personal pain tolerance, relentlessness, and a huge body of tacit knowledge passed down by apprenticeship.

[Dmitry (Dima) Lepikhin @lepikhin]

@andrewchen how much is the latter?