Tree Climber

swe intern @ BigCo looking for another internship
i tweet about AI/ML, finance

[Tree Climber @techtreeclimb]

if 3.5 Pro doesn't get GDM back into the top 3, the discourse on here is gonna be crazy...as of today they might not even have a top 5 model for coding

[Lentils @Lentils80]

🚨 Scoop: Some more internal Gemini 3.5 Pro checkpoints based on the new rev25 (revision) base appeared very recently

The internal ID of the most recent one is gemini-v4p1m-rev25-lovell, and it's still very bad at coding and hallucinates the knowledge cutoff date constantly

For reference, the old rev24 checkpoints from May were better

It's unclear whether 3.5 Pro will actually use the new base because it's still very undercooked and there isn't much time left

I attached an output from it below

[Tree Climber @techtreeclimb]

probably gonna see a lot of gdm slander on the TL in the coming days :(

[Noam Shazeer @NoamShazeer]

I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there.

It was a difficult decision to move on. I’m incredibly proud of the amazing team at Google and everything we’ve built together. It has been an honor and a pleasure to work with all of you.

[Tree Climber @techtreeclimb]

wonder if the Big 3 memory players also invest in oai at some point...seems likely

[Anthropic @AnthropicAI]

We've raised $65 billion in Series H funding at a $965 billion post-money valuation, led by @AltimeterCap, Dragoneer, @Greenoaks, and @sequoia.

This investment will help us advance our research and expand our capacity to meet growing demand for Claude.

[Tree Climber @techtreeclimb]

bigger than any neocloud?

[Tree Climber @techtreeclimb]

Ant initially paying SpaceX $1.25 bn per month thru May 2029, seems that the compute from expanded agreement announced today will be at a lower rate

[Tree Climber @techtreeclimb]

TL opinions on GDM model progress might be the worst since 2023

[Tree Climber @techtreeclimb]

also interesting that he was the one to close out the keynote

[Tree Climber @techtreeclimb]

Demis just now: Iso now in pre-clinical stage with multiple projects

[Tree Climber @techtreeclimb]

@jake_ayes @Lucknowcapital yeah ik what mgmt said but was trying to rationalize how they arrived at that number

upon closer reading it seems like they're just looking at the actual chip costs for capacity coming online in '26? not super clear

[Tree Climber @techtreeclimb]

@jake_ayes @Lucknowcapital I guess a majority of the ~$200bn is for non-AI stuff, e.g., logistics/warehouses

[Tree Climber @techtreeclimb]

the only Huawei mention:

[Tree Climber @techtreeclimb]

trying to read v4 paper as fast as I can

[Tree Climber @techtreeclimb]

🐋...can't believe it's finally here

[Scott Wu @ScottWu46]

here we go again

[Tree Climber @techtreeclimb]

cursor to xai based on all the vagueposting, i guess

[Tree Climber @techtreeclimb]

PLEASE WENFENG I NEED THIS

My open models are kinda lagging behind the frontier

I wanna help them out

[Jukan @ ICML @jukan05]

09:57:50 [Liang Wenfeng Reportedly Speaks Internally; DeepSeek V4 to Launch in Late April]

According to multiple sources speaking to media, DeepSeek founder Liang Wenfeng recently revealed in internal communications that DeepSeek's next-generation flagship model, DeepSeek V4, will officially launch in late April. Recently, DeepSeek's web interface quietly introduced two interaction modes — "Quick Mode" and "Expert Mode" — supporting concise search and long-form, complex problem-solving, sparking industry speculation about the V4 model release. Related reports indicate that DeepSeek V4 will not only feature trillion-scale parameters and a million-token-class context window, but will also for the first time achieve deep optimization for domestic Chinese chips such as Huawei Ascend, marking a critical step in China's AI industry toward "de-CUDAification." Foreign media reports suggest that in anticipation of the V4 launch, domestic tech giants including Alibaba, ByteDance, and Tencent have pre-ordered hundreds of thousands of next-generation AI chips, planning to offer DeepSeek's new model via cloud services and integrate it into their own AI products. As a result, prices of new AI chip products have risen approximately 20% recently. As of press time, DeepSeek has not responded to inquiries. (Sina Tech)

[Tree Climber @techtreeclimb]

seems v likely that they've passed oai

[Anthropic @AnthropicAI]

Our run-rate revenue has surpassed $30 billion, up from $9 billion at the end of 2025, as demand for Claude continues to accelerate. This partnership gives us the compute to keep pace.

Read more: https://t.co/XgSjL0And7

[Tree Climber @techtreeclimb]

@MetacriticCap Well, Jeff Dean is chief scientist of both GDM and Google Research, so I'd imagine that Google Research work still gets looked over by some GDM folks

Check out the Titans and ATLAS papers, also from Google Research; seems like GDM was okay with that stuff being published

[Tree Climber @techtreeclimb]

oai enterprise rev growing faster than consumer, will be ~equal EOY:
https://t.co/IAGFMVR287

[Tree Climber @techtreeclimb]

@MetacriticCap @theo_wayt probably just AI-related revenue, no? separated out from things like X ad revenue