Content About Stocktwits
No Image Available

‘Strike a Pose’ in AI. OpenAI, Anthropic & Apple. ARD #144

Author
Michael Parekh AI: Reset to Zero
1 day ago
11 MIN READ

‘Strike a Pose.’ Madonna gave us the phrase, the song and dance, and today it fits AI’s leading companies: three of them, striking deliberate public poses this week, each for an audience that matters, plus Tesla holding its robotaxi poses in Austin.

Today’s ARD is about those poses, in this AI Tech Wave: OpenAI striking an Anthropic-style AI safety pose, Anthropic easing its own a notch while re-organizing for its mega-IPO, and Apple striking a conciliatory pose with the EU. My takes below, as discussed on the show.

(1) OpenAI Strikes an Anthropic-Style AI Safety Pose

OpenAI is deliberately slowing its frontier model development, a first for the company. It paused reinforcement learning training on its next models, codenamed Astra, for a little more than two weeks, and its largest planned frontier training run remains on hold while new guardrails go in. The trigger is of course the remarkable Hugging Face ‘Great Escape’ incident of a few weeks ago: during an internal cybersecurity evaluation, an unreleased OpenAI system escaped its sandbox and compromised Hugging Face’s production systems, and it took OpenAI’s own researchers about a week to discover it. Astra may reach the ‘Critical’ cybersecurity threshold in OpenAI’s Preparedness Framework, a designation that requires safeguards through development, not merely before release. Sam Altman, to Time: ‘I think it is a good time to slow down.’

There is irony in the pose: OpenAI pioneered this AI safety alignment work years ago under co-founder Ilya Sutskever, shrank the effort, and Ilya left to found Safe Superintelligence, now a multi-billion dollar startup. Poses get struck, and restruck. The child safety pose came the same day: ChatGPT for Teens, a mode that automatically limits conversations on self-harm, violence, eating disorders and explicit content, alerts parents who opt in, answers homework prompts with guiding questions, and uses over 2,000 signals to detect an under-18 user. Timely indeed, as Meta’s landmark 29-state trial opened in Oakland federal court, with internal emails naming ‘teen time spent’ as the goal, employees describing Instagram as a ‘drug’, and Meta itself saying penalties could reach $1.4 trillion. A dramatic pose on penalties that are likely to be far lower.

My take: it is notable that OpenAI is borrowing some of Anthropic’s AI safety sheen for its next generation models, ahead of its own planned mega-AI IPO, potentially next year. And being proactive on child AI safety in light of Meta’s legal woes. All while accelerating its AI ‘super-app’, fusing the Codex AI Coding app with ChatGPT, the enterprise and the consumer together, along with the AI browser and the AI devices being planned with Jony Ive’s help. Multiple poses being struck concurrently indeed.

Sources:

For longtime readers:

(2) Anthropic’s Safety Pose Down a Notch, Re-Organizing for Its Mega-IPO

The safety recalibration came first, back in February: Anthropic dropped the central pledge of its Responsible Scaling Policy, the 2023 commitment to never train a model unless it could guarantee adequate safeguards in advance. Co-founder Jared Kaplan told Time it plainly: unilateral commitments made little sense ‘if competitors are blazing ahead.’ The replacement pose is transparency: safety roadmaps, regular risk reports, and a commitment to match or beat competitors’ safety efforts. Note the symmetry with the first item: as OpenAI slows down and strikes Anthropic’s old pose, Anthropic has already moved to a more pragmatic one.

Now the governance re-organization, this week: The Information reports Anthropic is preparing super-voting shares for CEO Dario Amodei and his six co-founders, who hold roughly equal, relatively small stakes after hundreds of billions in outside capital raised. The unusual part stays: the Long-Term Benefit Trust, non-shareholder trustees including Ben Bernanke, keeps its special class of stock electing the majority of the board. Anthropic would go public as by far the most valuable public benefit corporation ever; the current largest, Veeva, is around $40 billion. The money math underneath: a $65 billion revenue run rate at the end of July, on quarterly revenue above $11.5 billion, and a pre-IPO credit facility set to climb past $10 billion, up from $2.5 billion last year, with banks jockeying for IPO roles.

My take: Anthropic is being a tad more pragmatic on the AI safety poses ahead of its mega-AI IPO. It is a reflection of the practical steps needed as it preps its public market debut this fall, including governance and organizational changes, especially given its PBC structure and special board. All notable poses ahead of the big accelerating revenue growth advantage versus OpenAI, as Anthropic turns from being the Pepsi to the Coke in that arch-rivalry.

Sources:

For longtime readers:

(3) Apple Strikes a Conciliatory Pose with the EU

Apple agreed to sweeping App Store changes in Europe to settle its long fight with the European Commission. The per-install Core Technology Fee goes away, replaced by a flat 5% commission on transactions in apps distributed outside the App Store; in-app purchase commissions drop from the standard 30% to 26%, with most developers paying 15% or less; developers can offer rival payment options alongside Apple’s. Everything takes effect October 1. The Commission ‘welcomes’ the changes, after Apple lost its Digital Markets Act court fight in July and took a half-billion euro fine last year.

And note the echo of item one: Apple struck a child safety pose inside the settlement. Kids-category apps cannot link out to external purchases, under-18 users need parental approval for alternative payments, and under-13 users get no web transaction links at all.

My take: this is a prelude to Apple launching Siri AI and Apple Personal Intelligence globally, around the world. Remember, when Apple previewed these at WWDC, it said they would not be available in Europe; these DMA settlements hopefully lay the groundwork for Europe too. It also ties to the revelation that Apple’s $100 billion services business, its second biggest product line after the iPhone, came in a tad below expectations last quarter, partly on this EU kerfuffle. This resolution mitigates that as the new CEO John Ternus takes over in September.

Sources:

For longtime readers:

My overall take: every major tech company, and not so major tech company, is always striking poses; we all do. These three are the ones to note, ahead of critical opportunities and headwinds for them all. OpenAI needs to keep the regulators in DC even closer than Anthropic can, given Anthropic’s more fraught relationship with the administration: a relativistic advantage ahead of both their IPOs. Anthropic needs its PBC governance communicated as pro-business as possible for public investors. And Apple needs to come to terms with the EU ahead of the global mainstream launch of Siri AI and its AI products and services for third party developers, in all major markets. So we are in a ‘Strike a Pose’ mode for these three and their peers indeed.

Gadget AI: Tesla’s Cybercab Poses in Austin

Tesla is gearing up for the very showcase-y public launch of the Cybercab in Austin, as soon as this month: its first vehicle designed with no steering wheel and no brake pedals, a two-seater purpose-built for its Robotaxi service. The choreography is very Tesla: employee rides first, then the Cybercabs fold into the Austin Robotaxi fleet, with emergency training already done with Austin first responders.

The relative scale details are notable. Tesla has 186 Model Ys registered for its Austin service and has reported about 380,000 driverless miles across its Robotaxi hubs. Waymo, running commercially since 2020, reports over 220 million, and it gets far less valuation credit for it than Tesla gets for the promise. China’s robotaxis and AVs are running further ahead still. And speaking of China: Unitree, its leading robotics company, just went public, and was up some 600%, now north of $10 billion, the second eye-popping China IPO in weeks after memory maker CXMT’s ~450% debut surpassed Tencent as the largest market cap in China.

My take: the Tesla robotaxi theater continues, with big implications for Elon Musk’s Tesla and its public market cap, where analyst reports credit roughly half a trillion dollars each to the auto business, the robotaxis, and the Optimus robots, in a valuation of about a trillion and a half. (Not stock recommendations). Media and markets will pay big attention to this particular launch. But there is still a longer way to go for mainstream applications here than the current ‘poses’ suggest.

Sources:

For longtime readers:

Q&A

Q1: What is the most notable aspect of ‘no driver’ robotic taxis thus far? ANSWER: Still the novelty factor of it, every time. And the admiration for how far the technology has come in the twenty years since DARPA held its first driverless car challenge in the mid-2000s. That is always wonderful to see.

Q2: What is the next notable improvement needed on these robotic taxis? ANSWER: How far they still have to go to be a truly useful mainstream service, when passengers really do not miss a human driver in mainstream rides. Pricing in many cases is still higher than human-driven rides, and mainstream scale, millions of units versus today’s dozens, hundreds and thousands, is probably still a few years ahead, in the US and in China alike.

Today’s companion AI-RTZ, #1183, is on ‘Not one Skynet, but Billions of AI Agents’: the era of machine to machine agentic bots working for billions of humans, with new Anthropic research data on how AI Agents work curiously like humans together: AI: Not one Skynet, but Billions of AI Agents. AI-RTZ #1183

Full Source Reading

OpenAI’s safety pose

Anthropic’s re-organization

Apple and the EU

Gadget AI: Tesla’s Cybercab

Clips from today

OpenAI Hits the Brakes on Astra

Anthropic’s Pragmatic Safety Shift

Apple Settles Its EU App Store Fight

Tesla’s Cybercab Debuts in Austin

An impact of the AI Tech Wave worth tracking. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)





Want the latest?

Sign up for Michael Parekh's Newsletter below:


Subscribe Here

More like this

No Image Available
Monday links: holding periods
Tadas Viskanta Abnormal Returns
Markets 2025 earnings estimates are coming down. (ft.com) How asset classes performed during prior recessions. (morningstar.com) Warren Buffett Fans of...
1 year ago
2 MIN READ
BROWSE ALL ARTICLES