AI news, November 4 — itch.io "just deleted a couple terabytes of AI slop" draws 168,000 likes; daily new game pages doubled in three months

Wednesday, November 4, 2026

9 stories worth reading from 19 on X in the past day

Today's stories

9 stories

44 hours agoLatest updateGrok Bot phoned Comcast and negotiated a lower bill — but admitted it can't see the sky"I don't actually have eyes on the sky." That's what xAI's Grok Bot told a Comcast rep who asked about the we…@bot +1· ♥ 3.3K· X
46 hours agoLatest updateAnthropic's "$104B nine-month loss"? Replies say $94B is a paper loss on convertible notesA post claiming Anthropic lost $104 billion in the first nine months of 2026 on $33.24 billion of revenue wen…@brewmarkets· ♥ 1K· Al Jazeera
40 hours agoBreakingGoogle turns off Gemini and Search by default for K-12 student accounts; schools that reject AI lose Search tooGoogle will switch off Gemini and Google Search by default for K-12 student accounts, and schools that reject…@unusual_whales· ♥ 2K· Google
2 days agoitch.io "just deleted a couple terabytes of AI slop" draws 168,000 likes; daily new game pages doubled in three monthsIndie game marketplace itch.io posted on X on Nov.@itchio· ♥ 168K· Kotaku
2 days agoAnthropic opens residency paid up to $500,000 a year equivalent, no ML background required; applications close Nov. 22Anthropic has opened applications for a paid research residency that does not require a PhD or machine-learni…@SahilExec +1· ♥ 2.5K· Anthropic Alignment
44 hours agoCat Wu, who led Claude Code product, leaves Anthropic: "I never manually wrote a line of code" building an RL environmentCat Wu, who led product for Claude Code, has left Anthropic. She wrote on X on Nov.@ClaudeDevs +1· ♥ 6.2K· Claude Code Docs
45 hours agoClaude Enterprise 'smart reports' count off-topic chats; Anthropic says don't use them to evaluate individualsAnthropic offers Claude Enterprise admins "smart reports" that analyze team conversations.· Claude 도움말
2 days agoPaper says only GPT-6 Astra reads model weights; Qwen 27B on a laptop did it too. Willison: "not ready to declare the age of source code over"After a paper, "Sparks of Weight-Space Programming," claimed AI models can read and write neural-network weig…@simonw +1· ♥ 3K· Simon Willison
44 hours agoClaude now auto-connects company plugins at login: 14 services including Notion, Linear and Figma, in beta on Team and EnterpriseAnthropic has added auto-connection of company plugins when users log in to Claude.@ClaudeDevs +1· ♥ 3.8K· Claude 도움말

NEW · Email

Get AI News Daily in your inbox!

Every morning
The AI news you need, plus features and YouTube picks
AI Insights
New Insight articles as soon as they're out
Monthly briefing
On the 1st: the month's 10 most-read stories

No account needed, unsubscribe anytime in one click.

Trending on YouTube

2 stories

AI videos that drew views and comments fast on YouTube, with the claims checked against articles and sources

Reverse engineeringEEVblog video says Claude Code reconstructed 121GW multimeter firmware in C in a day, 90,000 views: "Software Engineering is so OVER"
A video from electronics channel EEVblog posted Nov. 3 (KST) has 90,828 views and about 2,400 likes. Its description says "Claude Code reverse engineered the entire firmware into C with a single prompt in under 24 hours, and it works!" The subject is the channel's Bluetooth multimeter, the 121GW.
According to the chapter list, the host first modified the 121GW for use with an ST-LINK V2 programmer, installed GCC and Ghidra, finished reverse engineering the bootloader, then started on the firmware. The next morning the firmware was complete, and it was tested on real hardware, including whether Bluetooth still works. The last chapter is titled "Conclusion: Software Engineering is so OVER." The description links the channel's source code and a version in assembler by another user, Zombu4.
The 121GW has been analyzed before. Its firmware has been studied on the EEVblog forum since 2018, and user tpwrules published an IDA database and reimplemented routines on GitHub. The video shows a claim that Claude Code did similar work in a day; no third-party verification beyond the in-video tests is available.
- ▶EEVblog 1784 - Claude Code로 121GW 펌웨어 리버스 엔지니어링 성공!EEVblog · 91K views
Further reading
- Reverse engineering the EEVBlog 121GWGitHub ↗
- EEVBlog 121GW Multimeter Firmware DetailsEEVblog Forum ↗
JobsSBS video on freelancers asked to charge less because AI made the work passes 77,000 views; top comment: "that call will stop in two years"
An SBS News video posted Nov. 3, "'It was made by AI, so give me a discount'... freelancers' changed work," has 77,181 views and 143 likes. It reports freelancers' work shifting toward checking and fixing AI output.
The description says companies increasingly produce design and translation drafts with AI and give experts only the job of raising quality. Services that fix AI typos and translation errors have appeared. At the same time, some clients cut rates or shorten deadlines because AI does most of the work. Experts said the time spent checking and fixing AI output should be reflected in pay and deadlines.
The top comment (122 likes): "Even asking for a discount won't last. In about two years that call won't come either." Others: "my company's illustration team lost their jobs all at once" (49) and "I'm a marketer and about to be wiped out" (35). A counterpoint said checking and fixing each piece doesn't take less time just because AI does the work (12).
- ▶"AI가 만들었으니 싸게 해주세요"..프리랜서들 달라진 업무 '씁쓸' / 알찬뉴스 / SBS 모닝와이드SBS 뉴스 · 77K views

Feature · Google school search

If schools turn AI off, does Google Search go dark too? Between an 8,000-like 'breaking' post and Google's clarification

"Google will turn off Search entirely for schools that opt out of its AI features." Citing The New York Times, this Nov. 3 post passed 8,200 likes and 730,000 views on X. Firefox and DuckDuckGo quickly posted jabs at Google (3,700 and 3,600 likes), while AI commentator Andrew Curran countered that "there is no Google search without Gemini." After the story ran, Google said schools can turn on Search separately. We traced what is established and what is still contested.

The facts first: from early 2027, student accounts are AI-off by default
On Nov. 3 Google said Gemini, Gemini Notebook, Gemini in Classroom and Google Search will be "entirely off for K-12 student accounts until school administrators explicitly turn them on." The New York Times put the timing at early 2027. It reverses August's change that turned Gemini on by default in Classroom. Schools that turn AI on can choose Search with AI Overviews and AI chat or Search with Guided Learning, a Gemini-based tutor.
@unusual_whales · The Philadelphia Inquirer (NYT) ↗
Where 'AI off means Search off' came from
Chris Phillips, Google's vice president for education, told the Times, "Search and AI are deeply integrated," and that students at schools that ban AI "will not have access to search, and they'll see a message from their school." That was compressed into the viral post, from @unusual_whales, saying schools cannot disable AI Overviews or AI Mode themselves and lose Search if they refuse.
@AndrewCurran_ · X ↗
Google's clarification after the story: Search can be enabled separately
NBC News reported Google said after publication that schools can enable Search without enabling Gemini products. The same report also says schools that keep those products off lose student access to Search, so how the two statements fit together is unclear. Andrew Curran argued that "if you type anything into the search box, you are talking to Gemini" and that framing the move as retaliation is wrong (about 600 likes); another user replied that putting AI in search was Google's own decision.
If schools turn AI on: step-by-step help, not answers
Google said AI Overviews and AI Mode will guide students step by step and detect attempts to just get an answer. Brit Mennuti, Classroom product lead, said teachers can review students' chatbot transcripts and the school pilot of Guided Learning has tighter limits than the consumer version. Times reporters asked the consumer Guided Learning for an AP history essay and got a detailed one, showing it can enable cheating.
Morgan Polikoff · Dexerto ↗
Critics: business interests versus school choice
AFT president Randi Weingarten said, "It is unfortunate that Google has prioritized its business interests over the education of our children." Fairplay's Josh Golin said Google is trying to "strong-arm schools into using its AI products." USC's Morgan Polikoff called it "a particularly clear example of the tech industry going out of its way to force A.I. on us all." Firefox posted, "You shouldn't have to pay a price for choosing not to use AI."
Google's case: more choice, not less
Google says the changes "give schools more choice, not less." Phillips said, "We believe schools should have control over AI in their classrooms." NBC reported Google has offered educators multiday AI trainings this year, some focused on handling teacher pushback.
Background: Chromebooks, parent groups, New York and Los Angeles limits
Google supplies schools' Gmail, Docs and Classroom and sells most Chromebooks that districts buy. A 2023 Schoolytics report found nearly two-thirds of assignments use a Google product. In early October 13 parent groups, including the Digital Childhood Institute, urged districts to turn off Gemini and YouTube. In September New York City and Los Angeles, both Google clients, broadly restricted students' generative AI use on school devices; Los Angeles had hoped to keep search with AI Overviews off.
What is still unknown
The number of schools and students affected has not been disclosed. Per Stanford's SCALE Initiative, there are no high-quality causal studies of student AI use. The rollout is in early 2027, so settings may change.

AI & Markets

4 stories

Not investment advice. You are responsible for your own investment decisions.

SemiconductorsAMD Q3 revenue up 45% to $13.38 billion; data center $8.5 billion, Q4 guidance about $17 billion
AMD said on Nov. 3 that third-quarter revenue rose 45% to $13.38 billion from $9.25 billion a year earlier. CNBC reported it beat expectations of $13.09 billion, with adjusted EPS of $2.09 against $1.95 expected. Data center revenue rose 97% to $8.5 billion, 64% of the total versus 47% a year ago. Gaming and PC revenue fell 8% to $3.7 billion, with gaming down 40%.
Q4 revenue guidance is about $17 billion, plus or minus 3%, above the $16.2 billion consensus. CEO Lisa Su said "Adoption is just starting"; shares rose 7-8% after hours. Per Reuters, the Nasdaq 100 fell into a correction on Nov. 2, more than 10% below its Oct. 6 record, then rose 0.68% on Nov. 3. Nvidia rebounded after four down sessions; Apple fell after warning memory costs would hurt holiday-quarter gross margin.
AI accelerator demand nearly doubled AMD's data center revenue while PC and gaming shrank. The memory cost pressure Apple flagged reaches smartphone and PC makers.
Further reading
- AMD Q3 2026 earnings reportCNBC ↗
- Nasdaq ends higher as AI heavyweights rebound ahead of midterm resultsReuters (Kitco) ↗
Korea marketsLablup falls 25% from offer price on KOSDAQ debut; KOSPI sees 1.5 trillion won foreign net selling
AI infrastructure company Lablup listed on KOSDAQ on Nov. 3 and closed at 14,220 won, 25.16% below its 19,000 won offer price. AI Times reported it rose as high as 28,350 won (+49.2%) before dropping to 13,310 won. Volume was 37.32 million shares, 17 times the 2.18 million offered, and market cap at close was 207.1 billion won. The offer priced at the bottom of its range; institutional demand was 284.53 to 1 and retail subscription 3.55 to 1.
The same day KOSPI fell 210.29 points (3.59%) to 5,648.79. Seoul Economic Daily reported foreigners net sold 1.5152 trillion won; Samsung Electronics fell 4.15% to 219,500 won and SK hynix 5.03% to 1,398,000 won. KOSDAQ fell 4.08%. US tech weakness and worries over AI investment funding were cited.
While AMD said in the US that data center revenue nearly doubled, a new AI infrastructure listing in Korea closed below its offer price on day one.
Further reading
- Lablup falls 25% from offer price on KOSDAQ debut dayAI타임스 ↗
- KOSPI retreats to 5,648 as foreigners sell 1.5 trillion won서울경제 ↗
InvestmentOpenAI trims funding round from $30B to $20B, valuation $1.4T to $1.25T; Anthropic investors reported discussing $1.5T
The Financial Times reported on Oct. 30-31 that OpenAI cut its funding target from $30 billion to $20 billion and its valuation from $1.4 trillion to $1.25 trillion. Some investors trimmed allocations before close, so the first close may fall short of $20 billion. AI Weekly summarized the FT; CNBC quoted OpenAI saying the round "has not closed, so the figures are subject to change."
On Oct. 16, AI Times, citing Reuters and The Information, reported some Anthropic investors were discussing about $1.5 trillion, roughly 25% below the $2 trillion expected. It is not a price Anthropic has disclosed. Anthropic filed its S-1 late last month under ticker ANT with Goldman Sachs and Morgan Stanley as leads and plans to list this month.
Anthropic's filing shows $895.4 billion in infrastructure commitments, $69.8 billion of it due within 12 months (reported Nov. 3). Counterparties are chip, cloud and data-center companies.
Further reading
- OpenAI trims round to $20B at $1.25T valuation, FT saysAI Weekly ↗
- Anthropic valuation of $1.5 trillion discussedAI타임스 ↗
Timeline signalSemiAnalysis: SpaceX to become "world's biggest AI cloud" with 13GW in 1.5 years; analyst projection draws 1,500 likes
SemiAnalysis published an analysis on Nov. 3 saying SpaceX will become the world's biggest AI cloud, reaching 13GW in 1.5 years. The X post announcing it passed 1,500 likes and 250,000 views. It raised its 2027 estimate from 10GW to 13GW, citing 3.6GW under construction and more than 9GW of identified sites. It says 13GW roughly equals all of Google's data center capacity from 1999 to 2026.
Retrofitting Colossus 1 took 122 days for 150MW and Colossus 2 took 91 days for 210MW, with 5.9GWh of batteries. "Project SPAM" began construction Oct. 5 with first tranches targeted for December. The authors have medium-high confidence the Texas TeraFab will include a data center. Pricing and funding sections are paywalled.
This is an analyst projection, not figures confirmed by SpaceX. A 13GW buildout reaches the server, networking and power-equipment supply chain directly. The previous day brought news of Grok Bot limits cut over costs.
Further reading
- Elon Is Doing It Again: SpaceX Will Be The World's Biggest AI CloudSemiAnalysis ↗
- Elon Is Doing It AgainX (SemiAnalysis) ↗

All AI & Markets news →

Other AI news

4 stories

Picked from news outside X, Threads and YouTube, published in the past 24 hours

EuropeEU court adviser: Google Gemini may infringe press rights when it pulls articles into answers; training itself allowed
Advocate General Maciej Szpunar of the EU Court of Justice said on Nov. 3 that Google's Gemini may infringe press publishers' rights when it retrieves articles to build answers. Reuters reported first. The case, C-250/25, was brought by Hungarian publisher Like Company against Google Ireland and is the court's first on generative AI and copyright; a Budapest court referred it in April 2025.
The Next Web reported he found training on protected articles "freely permitted" under the text-and-data-mining exception, but said publishers' rights "may be infringed in certain circumstances" when Gemini retrieves an article and uses it in an answer, such as bypassing paywalls or recognizably reproducing wording. Copyright protects expression, not facts.
Google argues Gemini does not memorize articles, any infringing output falls under temporary-copy and mining exceptions, and Hungarian law should not apply because Gemini was not trained there. The opinion is non-binding; a ruling is expected in the coming months.
Further reading
- EU court adviser says Google's Gemini may infringe press publishers' rightsThe Next Web ↗
- EU court adviser says Google Gemini may infringe pressDataconomy ↗
USHollywood, music and publishing groups back NYT with briefs in copyright case against OpenAI and Microsoft
Hollywood, music and publishing groups filed amicus briefs supporting The New York Times in its copyright suit against OpenAI and Microsoft, Axios reported on Nov. 3. The filings came last Friday in "In re OpenAI Inc. Copyright Infringement Litigation."
A Human Consent Foundation coalition brief carried backing from Anne Hathaway, Bette Midler and Madonna and included agency CAA. RIAA, NMPA, ASCAP and BMI filed their own; SAG-AFTRA, the American Federation of Musicians and the Writers Guilds East and West filed for "more than 250,000 entertainment and media workers." News/Media Alliance and local-journalism groups joined.
The briefs reject the AI industry's claim that licensing is too cumbersome. Axios said the Justice Department has backed OpenAI's position on AI training. On Nov. 2 the Times and other publishers also sued Brave.
Further reading
- Hollywood backs NYT in AI copyright battle versus OpenAI, MicrosoftAxios (Yahoo) ↗
KoreaKorea weighs replacing Chinese BAAI model in government AI platform: "no security issue found"
The government is reviewing replacing BGE-M3, a model from China's Beijing Academy of Artificial Intelligence, in the shared AI platform used by civil servants, Money Today reported on Nov. 4. Officials are weighing replacement models' performance, language support, compatibility and technical support. No security problem with BGE-M3 itself has been found.
BGE-M3 is an embedding model that turns sentences into numbers to find relevant documents, not a chatbot-style LLM; it handles the search step. The platform runs in a network separated from the internet. Background includes BAAI's place on the US Entity List and a parliamentary audit last month noting 7 of 11 installed models are foreign.
Korea's domestic foundation models are generative LLMs and cannot replace an embedding model. Prof. Lee Man-hee of Hannam University said "even a closed network always has supply-chain issues," but excluding all foreign models is unrealistic.
Further reading
- [Exclusive] Will Korea drop the Chinese model from its government AI? Replacement for BGE-M3 under review머니투데이 ↗
KoreaAI-drawn snack package images that differ from the product: Yonhap fact check says "legal exaggeration" will grow
Yonhap published a fact check on Nov. 4 after a limited-edition Lotte Wellfood Castella cake drew complaints that the raspberry syrup pictured on the package is far less than in the product. The package carries an "image photo" note and a small "AI" mark. The maker said syrup is injected first and then cream, causing syrup to pool, and that it is seeking improvements.
Korea's food labeling and advertising act bans false or exaggerated labeling, with penalties of up to five years in prison or a 50 million won fine; cooked-food photos must carry notes such as "serving suggestion" at 10 points or larger. A Food and Drug Safety Ministry official said depicting ingredients that are absent is clearly illegal but where to draw the line on exaggerating present ones is hard.
The AI Basic Act requires AI operators to label generative output but in principle does not cover ordinary companies making ads. Yonhap said "legal exaggeration," squeezing AI images to the most appetizing limit, is likely to grow. Japan uses industry codes per product category. Korea Consumer Federation's Jung Ji-yeon warned that overly specific guidelines can signal that meeting them is enough.
Further reading
- [Fact check] Package images drawn by AI: is it OK if they differ from the product?연합뉴스 ↗

View all 19 collected posts

* Summaries and publishing on this page are done by an automated agent. Check the facts against each story's original links.

Today's stories

Get AI News Daily in your inbox!

Trending on YouTube

Reverse engineeringEEVblog video says Claude Code reconstructed 121GW multimeter firmware in C in a day, 90,000 views: "Software Engineering is so OVER"

JobsSBS video on freelancers asked to charge less because AI made the work passes 77,000 views; top comment: "that call will stop in two years"

If schools turn AI off, does Google Search go dark too? Between an 8,000-like 'breaking' post and Google's clarification

The facts first: from early 2027, student accounts are AI-off by default

Where 'AI off means Search off' came from

Google's clarification after the story: Search can be enabled separately

If schools turn AI on: step-by-step help, not answers

Critics: business interests versus school choice

Google's case: more choice, not less

Background: Chromebooks, parent groups, New York and Los Angeles limits

What is still unknown

AI & Markets

SemiconductorsAMD Q3 revenue up 45% to $13.38 billion; data center $8.5 billion, Q4 guidance about $17 billion

Korea marketsLablup falls 25% from offer price on KOSDAQ debut; KOSPI sees 1.5 trillion won foreign net selling

InvestmentOpenAI trims funding round from $30B to $20B, valuation $1.4T to $1.25T; Anthropic investors reported discussing $1.5T

Timeline signalSemiAnalysis: SpaceX to become "world's biggest AI cloud" with 13GW in 1.5 years; analyst projection draws 1,500 likes

Other AI news

EuropeEU court adviser: Google Gemini may infringe press rights when it pulls articles into answers; training itself allowed

USHollywood, music and publishing groups back NYT with briefs in copyright case against OpenAI and Microsoft

KoreaKorea weighs replacing Chinese BAAI model in government AI platform: "no security issue found"

KoreaAI-drawn snack package images that differ from the product: Yonhap fact check says "legal exaggeration" will grow