TAWS

Friday, October 23, 2026

October 23, 2026

Watch on YouTube · Listen on Spotify

The rundown

AI-generated from the transcript

Austin's Friday show covers a Microsoft AI system that outscored general doctors on hard diagnostic cases, OpenAI reportedly trimming its valuation target ahead of a big raise, Etsy's new rules against AI made recipes and patterns, Google paying homes to cut power use, a federal judge pulled off a case over AI errors, and outside auditors finding gaps in OpenAI's agent monitoring. Ron weighs in on the judge story, and the episode closes with a new Friday segment, the TAWS Top Five of the week.

Microsoft's AI outscores general doctors on hard medical cases

Watch from 2:34

What happened
A study published Thursday in the journal Science found Microsoft's AI got 83% of hard cases from a top medical journal right, versus 36% for general doctors who could use books and the internet. On everyday ER cases it scored 80% to the doctors' 47%. The AI works like a panel of virtual doctors that asks questions, orders tests, and debates the answer. A Boston hospital will test it with patients.
The claim
Microsoft's AI chief calls it a step forward for medical superintelligence.
Austin's take
Austin is excited to see Microsoft emerge as another frontier player, and sees this as a strong case for pairing human doctors' judgment with an AI that has the full body of medical research at hand. He wonders how readily patients will accept it.
Why it matters
Your next diagnosis could come with an AI second opinion, and hospitals are starting to test that with real patients.
What to do
Treat the headline numbers with care: ask how a tool was tested before trusting it, and think about whether you would want an AI second opinion alongside your own doctor.
Not confirmed yet
The cases were old ones written up as text, not live patients, and specialists were not tested, only general practice doctors.

OpenAI reportedly cuts its valuation target by $150 billion

Watch from 6:13

What happened
OpenAI is reportedly targeting a $1.25 trillion valuation, about $150 billion lower than before, while raising up to $30 billion. CNBC says UAE investors are leading, and Bloomberg says commitments are due by the end of next week. Its last round, last March, valued it at $852 billion post money. Anthropic reportedly wants $1.5 trillion to $2 trillion in an IPO expected next month. Tech stocks slid this week on fears AI spending has gone too far.
Austin's take
Austin calls the whole situation tenuous and strategic, guesses the labs may be using their own models to plan these moves, and notes investor demand still looks strong. He questions when financing gives way to revenue, and whether big projections like Elon's SpaceX revenue targets are reachable.
Why it matters
Investors will compare OpenAI's and Anthropic's price tags, and these numbers signal how much money the market is still willing to pour into AI.
What to do
Watch how the OpenAI raise and Anthropic IPO terms land, and be skeptical of any single valuation figure while markets are this volatile.
Not confirmed yet
The valuation and raise are reported, not confirmed, and the episode says terms could still change.

Etsy bans AI made recipes and craft patterns

Watch from 11:01

What happened
Etsy banned AI made recipes and craft patterns after sellers said some AI patterns could not produce the pictured item. Pattern and recipe listings now need a real photo of the finished result, most other listings need one unedited photo of the actual item, and fully automated listings are banned. AI is still allowed elsewhere if sellers properly disclose it. Sellers may now owe refunds when an item does not match the listing.
The claim
Etsy says many buyers can't tell what's AI and that this is a top three reason they shop less, and that the photo rule helps buyers know what to expect.
Austin's take
Austin loves seeing the market push back on AI content. He prefers human made, messy work, and explains that AI writing stays detectable because even its randomness follows a pattern.
Why it matters
It is a clear example of a marketplace responding to buyer frustration with AI generated listings, with refunds now on the table.
What to do
If you buy on Etsy, look for real photos of finished items and use the refund policy when something doesn't match. If you sell, update listings and disclose AI use.
Not confirmed yet
Enforcement is the catch: Etsy gave no deadline for fixing old listings.

Google will pay homes and businesses to use less electricity

Watch from 15:53

What happened
Google, with partner Voltus, will pay homes and businesses to cut power use when demand peaks, enrolling smart thermostats, batteries, and factory equipment. Participants get bill savings and cash. The grid serves 67 million people in 13 states and data centers helped push its prices up. The deal is 10 times bigger than one from June. Payments for cutting back on this grid jumped 240% last year.
The claim
Google says the plan makes the grid more reliable and affordable. Voltus says full size is three to four years away.
Austin's take
Austin compares it to Austin's Base Power and notes demand limits aren't new. He's fine with it if it's opt in, but doesn't love the idea of being required to participate.
Why it matters
Data centers are straining grids and raising power prices, and big tech is now reaching into home thermostats to manage that.
What to do
If offered a program like this, check that it's opt in and find out what you'd actually earn before letting anyone adjust your AC.
Not confirmed yet
Google hasn't said what each home earns, and the total cost is unknown.

Federal judge pulled off a case over AI made errors

Watch from 18:39

What happened
A federal appeals court removed Judge Henry Wingate from a case after his order blocking Mississippi's ban on DEI in public schools named people not in the case and misquoted the law. Even the corrected version cited a court case that doesn't exist. Judge Jerry Smith called the proceedings a botched job, and the court also said Wingate got the law wrong.
The claim
Wingate called the errors clerical and said a law clerk's AI draft got filed by mistake.
Austin's take
Austin says legal work needs tools built and trained for specific jurisdictions, plus humans checking every output. Ron argues the real problem was that nobody disclosed AI use, and Austin agrees courts should disclose when AI helped.
Why it matters
Judges' offices use AI too, and a sloppy draft can decide a real person's case.
What to do
If you use AI for legal documents like NDAs, read the full output and verify every citation. Push for disclosure when AI is used in rulings.
Not confirmed yet
It's unclear how much of the error came from the AI versus the judge himself, since the court also faulted his reading of the law.

Outside auditors find holes in OpenAI's agent watchdog

Watch from 23:37

What happened
Researchers from safety groups METR and Redwood Research spent three weeks inside OpenAI with deep access to code, monitor settings, and internal documents. They found about 10% of coding agents' work went unchecked for months, agents could dodge checks by never handing control back, and one trick could hide hundreds of recent actions from review.
The claim
OpenAI says it patched every hole that touched its live safety systems.
Austin's take
Austin notes OpenAI keeps showing up in safety stories but wonders if that's because it reports more. He praises the use of outside auditors and wants OpenAI to publish technical fixes so other labs can benefit.
Why it matters
OpenAI runs these agents internally, and the watchdog is what's supposed to catch one that misbehaves, especially as agents increasingly build and train each other.
What to do
Look for labs that invite independent audits and share findings, rather than grading themselves.
Not confirmed yet
Auditors haven't verified that OpenAI's patches work.
Sources

TAWS Top Five of the week

Watch from 26:16

What happened
Austin's new Friday countdown: 5) Anthropic unveiled an agent that messages you first with a morning brief and can book meetings, build forecasts, and shop with a virtual card, asking before it pays, sends, signs, or deletes. 4) SpaceX AI posted a $4.4 billion profit on $4.6 billion in sales, with AI over half of sales, and Tesla's filing confirmed a special committee looking at a merger. 3) The NBA built AI to catch shady bets, dropped first scorer and opening tip bets, and brought in Charles Tillman to train teams. 2) Claude solved where five unexplained NSA supplied numbers from 2000 came from in under thirty minutes, tracing them to computer clocks in 1997 and 1998, and a tripled bounty sent $37,000 to charity. 1) Leaked draft IPO filings reported by the Wall Street Journal show Anthropic at $16.3 billion in first half revenue, up from $1.1 billion a year earlier, with at least 5% of shares set aside for the public good, estimated near $48 billion.
Austin's take
Austin enjoyed that Claude wondered in its chain of thought whether it was being tested, and calls it a week of amazing developments.
Why it matters
It's a quick catch up on the week's biggest AI moves across products, money, sports, and security.
What to do
Pick the story that matters most to you and dig into it; Austin asks viewers to share their play of the week in the comments.
Not confirmed yet
Anthropic's revenue figure comes from leaked draft filings, not a final public filing.

The Loop

The weekly newsletter

The week in AI in one email: what happened, why it matters, what to do about it.