varunsingla.com

← All entries

Day 232 5 min read · PDF

Science Gets an AI Co-Researcher, ChatGPT Learns to Make GIFs, and the Compute Race Hits Wyoming

Anthropic's study of 207,049 research sessions shows how scientists really use AI, OpenAI's Images 2.6 adds animated GIFs and reusable reference images, and Google's $17 billion Wyoming data centre plus OpenAI's $1.25 trillion-valuation funding show the physical and financial cost of the race.

Infrastructure & Economics Models & Frontier

Themes: Models & Frontier · Infrastructure & Economics Note: the web-search tool returned only search-result summaries today and direct page fetches were blocked. Dates, figures and product names are secondary reports from news and aggregator sites and are unverified; check the original announcements.

Part 1How Scientists Actually Use AI: Anthropic's 207,049-Session Study

Reports dated November 5, 2026 say Anthropic analysed 207,049 research sessions carried out with Claude and Claude Code to see how AI is used in science. Researchers mainly use it for data analysis and writing. The study estimates that research output could roughly double in some fields if current usage spread widely. A separate report says the pace of improvement in Anthropic's models on certain task types is about twice as fast as a year ago. In the same news cycle, OpenAI reportedly released BioBasicsBench, a 100-task benchmark testing whether AI agents can interpret raw data from laboratory experiments. The concept, explained simply: A usage study looks at what people actually do with a tool, instead of what a lab says the tool can do. Think of it as watching how people really use a power drill rather than reading the box. Most science time is spent on the unglamorous parts, such as cleaning data, writing code to analyse it and drafting papers, so even a modest AI speed-up there adds up. A benchmark such as BioBasicsBench is a fixed exam: the same 100 problems posed to every agent so scores can be compared. Why it matters: The claim that output could double is an estimate about adoption, not a measured result, so treat it as a scenario. Still, it shifts the debate from 'can AI make discoveries' to 'how much routine work can it remove', and it shows which skills (data handling, coding, writing) scientists lean on AI for first.

Part 2ChatGPT Images 2.6: GIFs, Reference Images and 4 Billion Images a Week

OpenAI launched Images 2.6 on November 4, 2026, adding native animated GIF creation to ChatGPT, Codex and the API. You can make a GIF from a text prompt or a photo, refine it with follow-up instructions, change playback speed, then save or share it. Reusable reference images let you keep pictures of yourself, a pet or a character ready for new creations. OpenAI says images generate about 50% faster and that about 4 billion images are made weekly across its products. ChatGPT cannot yet edit a GIF you upload, and the system card reportedly admits that action-related policy violations are hard to detect. Sources disagree on which plans get GIFs. The concept, explained simply: A GIF is a short looping animation: a handful of images shown one after another. An image model that makes GIFs must keep the subject looking the same from frame to frame, which is called consistency. Reference images solve a related problem: instead of describing your character in words every time, the model is shown the picture, so it can reuse the same face or mascot. Why it matters: Consistency is the missing piece that turns image generators from toys into production tools for memes, brand mascots and storyboards. The safety note matters too: a still image is easy to check, but motion adds actions that are harder for filters to judge.

Part 3Compute and Capital: Google's Wyoming Data Centre and OpenAI's

Google reportedly announced on November 6 a $17 billion investment in its first data centre in Laramie County, Wyoming. The same day Bloomberg reported that OpenAI completed an $11 billion portion of a funding round at a $1.25 trillion valuation, below the level it had earlier sought, and OpenAI's CFO said the company will keep expanding US data centres despite election-season pressure to limit some projects. Separate commentary estimated OpenAI and Anthropic could take 25-40% of new global AI compute in 2027. AI Learning -- Day 226 | 2026-11-09 | Page 2 The concept, explained simply: A data centre is a warehouse of servers whose cost is dominated by chips, power and cooling. A valuation is the price investors implicitly put on the whole company in a funding round. Because training and serving models needs enormous computing capacity, AI labs raise money first and spend it on chips and buildings, so funding news and data-centre news are two views of the same pipeline. Why it matters: Where the compute is built decides who can train frontier models and how much electricity local grids must supply. A valuation below the one a company sought is a small sign that investors are pricing more carefully, which matters as Anthropic's IPO approaches. VIRAL / OPEN-SOURCE SPOTLIGHT: FlickReels (AI short-drama app) FlickReels is a mobile app for watching and making short AI-generated drama stories in the vertical, episode-by-episode style of micro-dramas. Reports say it jumped 20 places this week to #6 in the App Store Entertainment chart, following a similar climb by VibeShort in October. What it does: you watch (or generate) episodes of a few minutes each where scripts, characters and visuals are produced largely by AI. Why it is taking off: short-drama formats already had a habit-forming audience, and AI cuts production cost so sharply that apps can release far more episodes. It also raises the same question as the viral commission-reuse video this week: who owns and gets paid for the art that AI is trained on or animates? ChatGPT's new GIF tool is the mainstream cousin of the same trend.

Market signal

Bloomberg reportedly says investors weighing Anthropic's IPO are discussing a $1.2-1.3 trillion valuation, while OpenAI closed $11 billion at $1.25 trillion. The two top labs are being valued almost level, and both are somewhat below earlier asks. The signal: capital is still abundant, but pricing discipline is arriving as public markets near.

Practical takeaways
Use AI first on the boring 80%.

The study shows scientists lean on AI for data analysis and writing, so start there before asking it for breakthroughs.

Treat 'could double' as a scenario.

Adoption estimates are forecasts, so ask what usage level they assume before quoting them.

Save reference images for repeat characters.

A reusable picture keeps a mascot or persona consistent across GIFs and posts.

Check licences before animating other people's art.

Commissioned or scraped artwork may not cover derivative animation.

Watch funding and chips together.

A valuation tells you the money raised and a data-centre announcement tells you where it is being spent. SELF-TEST Q1. What was studied in Anthropic's November 5 report? A1. 207,049 research sessions with Claude and Claude Code, mostly data analysis and writing. Q2. What does a reference image do in Images 2.6? A2. It lets you reuse a person, pet or character across new creations without re-uploading or re-describing it. Q3. Why are funding and data-centre news linked? A3. Labs raise money and spend it on chips, power and buildings to train and serve models. Q4. What is a valuation? A4. The price investors implicitly put on the whole company in a funding round.

Self-test
What was studied in Anthropic's November 5 report?

207,049 research sessions with Claude and Claude Code, mostly data analysis and writing.

What does a reference image do in Images 2.6?

It lets you reuse a person, pet or character across new creations without re-uploading or re-describing it.

Why are funding and data-centre news linked?

Labs raise money and spend it on chips, power and buildings to train and serve models.

What is a valuation?

The price investors implicitly put on the whole company in a funding round.

What we're watching · tracker

whether Anthropic's IPO marketing begins this week and at what price range. What We're Watching: OpenAI or Anthropic primary posts confirming today's reported figures. What We're Watching: whether ChatGPT GIF editing of uploaded files arrives. What We're Watching: more AI short-drama apps entering the App Daily AI Learning for Varun AI Learning -- Day 226 | 2026-11-09 | Page 3 Store charts.

Sources. aiweekly.co (Images 2.6 alert, AI short dramas), techrepublic.com, runtimewire.com, digit.in and letsdatascience.com (Images 2.6); aidapted.ro, note.com, jmacweb.com and kucoin.com news summaries (Anthropic study, OpenAI funding, Google Wyoming, BioBasicsBench). Figures are secondary reports and unverified. Daily AI Learning for Varun
Related entries
VS
Varun Singla
Singapore · About · Learning in public