Choblin

AI Insights - AI Updates - Exclusive

[Choblin @choblin29]

🚨 EXCLUSIVE NEWS: A new model name has surfaced in Google's own SDK activity: gemini-3.7-flash.

The commit is titled: 'feat: Add gemini-3.7-flash to ModelOption enum.'

In other words, Google was adding gemini-3.7-flash to the SDK's generated list of recognized model options. It has not landed on main yet.

The precedent is interesting: gemini-3.6-flash was added to this same generated model registry on July 21 - the exact day Google released Gemini 3.6 Flash.

And there is another strange piece of the timeline. More than a month before 3.6 Flash launched, Google had already wired gemini-4-flash-preview into a separate tokenizer subsystem on June 10.

So Google had Gemini 4 Flash Preview appearing in SDK infrastructure before it publicly shipped Gemini 3.6 Flash, and now gemini-3.7-flash is appearing as another model option.

I increasingly think Google may be developing the 3.x Flash line and Gemini 4 in parallel.

[Choblin @choblin29]

https://t.co/pMczN1nPBZ

[Choblin @choblin29]

🚨BREAKING: BANK OF AMERICA plans to deploy $250 BILLION into U.S. infrastructure.

>Data centers + computing
>Power generation
>Energy storage
>Transportation
>Natural gas

Just 2 days ago, Morgan Stanley announced its own $1.5 TRILLION infrastructure initiative.

Wall Street is going all-in on the physical AI buildout

[Choblin @choblin29]

https://t.co/JILL4lKbdR

[Choblin @choblin29]

South Korea just put Moon landings, quantum computers, fusion power, brain-computer interfaces and space data centers into one national technology roadmap.

The new 7-project “SEED” strategy targets:

>Moon landing by 2030, second lander in 2032
>independent LEO communications network by 2035
>domestic error-corrected 100-qubit quantum processor by 2029
>fusion-based electricity generation in the late 2030s
>AI-bio infrastructure by 2030
>commercial brain-computer interfaces by 2035
>AI drug discovery and autonomous laboratories
>technologies related to space data centers
>$7.1 billion for critical materials and supply-chain technologies by 2030

This is basically a country-level moonshot stack for the 2030s.

Absolutely insane ambition.

[Choblin @choblin29]

Chief Master Sgt. of the Air Force David R. Wolfe announced it during a keynote on enlisted professional military education. The Air Force release gives no model, vendor or rollout date.

[Choblin @choblin29]

The U.S. Air Force is developing an AI agent to help Airmen understand applicable policies and career opportunities, so human development advisers can focus on cases requiring deeper analysis.

[Choblin @choblin29]

https://t.co/X4JLPzV5PL

[Choblin @choblin29]

🚨Exclusive: Google is planning to release Gemini 3.5 Pro+ instead of Gemini 3.5 Pro.

Four days after SemiAnalysis reported that Google had silently cancelled the model, fresh Google-generated GenAI SDK code still explicitly uses 3.5 Pro+ as the boundary for a new AGENTIC media-processing system. The latest revision was pushed to GitHub about six hours ago.

Importantly, 3.5 Pro+ may not be a separate “Gemini 3.5 Pro Plus” model. Google may appear to be using the + to mean Gemini 3.5 Pro and newer models.

Older models use fixed-rate frame extraction. From 3.5 Pro onward, Google says the default becomes “model-driven dynamic navigation,” allowing the model itself to dynamically navigate media.

Fresh Google code mentioning the model generation makes reports of its cancellation considerably less certain.

And we are also getting Gemini 3.7 Flash very soon.

[Choblin @choblin29]

🚨EXCLUSIVE NEWS: Is Google preparing Gemini 3.5 Pro+ instead of Gemini 3.5 Pro?

Fresh Google GenAI SDK code pushed about six hours ago says older models use fixed-rate frame extraction, while 3.5 Pro+ defaults to a new AGENTIC mode with “model-driven dynamic navigation.”

Google has only announced Gemini 3.5 Pro. The + may simply mean 3.5 Pro and newer models, but given recent reports that 3.5 Pro was cancelled, the timing is interesting.

[Choblin @choblin29]

🚨EXCLUSIVE NEWS: Is Google preparing Gemini 3.5 Pro+ instead of Gemini 3.5 Pro?

Fresh Google GenAI SDK code pushed about six hours ago says older models use fixed-rate frame extraction, while 3.5 Pro+ defaults to a new AGENTIC mode with “model-driven dynamic navigation.”

Google has only announced Gemini 3.5 Pro. The + may simply mean 3.5 Pro and newer models, but given recent reports that 3.5 Pro was cancelled, the timing is interesting.

[Choblin @choblin29]

OH MY GOD 😭🔥

Free users are absolutely winning right now.

“Think with GPT-5.6 Luna” is seriously incredible. The reasoning quality is on another level.

And now free users basically have unlimited access to GPT-5.5-level intelligence.

WOW.

Can Dario do this with Sonnet 5?

[Choblin @choblin29]

https://t.co/kAmSIzzDpc

[Choblin @choblin29]

🚨 BREAKING: Former OpenAI CPO Kevin Weil is reportedly aiming to raise $150M for a new AI science startup at a valuation of at least $750M.

The startup has been pitched as a way to gather scientific data for AI models.

AI-for-science is becoming the next founder gold rush

[Choblin @choblin29]

https://t.co/7dm8bikMba

[Choblin @choblin29]

CoreWeave reported $2.58 billion in Q2 revenue as demand for AI compute continues to grow. Its technology and infrastructure expenses simultaneously rose 125% to $1.51 billion.

[Choblin @choblin29]

https://t.co/aYqDDiOKmF

[Choblin @choblin29]

Supermicro expects $65 billion to $72 billion in revenue for fiscal 2027, versus Wall Street's $52.5 billion estimate. The company is betting that demand for AI servers will drive another year of growth.

[Choblin @choblin29]

The Claude watermark thing creates an absurd authorship problem:

You write a 3,000-word article yourself.

Claude fixes 6 typos.

Now it can carry an AI signal just like text Claude wrote from scratch.

“AI wrote this” and “AI touched this” are VERY different claims.

[Choblin @choblin29]

https://t.co/n5PKjrLoJ7

[Choblin @choblin29]

🚨 Spotify is about to draw a VERY visible line between human artists and synthetic ones.

Starting mid-September:

>AI-generated artist identities get an “AI Persona” badge
>badges appear on profiles, Search AND playlist track rows
>Spotify can identify them itself even if the artist doesn’t disclose
artists can appeal
>listeners will eventually be able to report suspected AI Personas

And the big one:

Spotify says AI Personas will be excluded by DEFAULT from editorial + algorithmic recommendations.

This is no longer just “label the AI.”

Spotify is now separating synthetic identities from human artists at the DISTRIBUTION layer.