Checking the Receipts: June

Round three. Same rules as April and May: dig up what tech leaders said about AI in a given month, hold it against what actually happened. 🍷 aged like wine, 🥛 aged like milk, 🤷 jury’s still out.

June is Apple month, thanks to WWDC. That makes it reliably rich for this series; 2024 was an especially good vintage.

2022

Pre-ChatGPT. June 2022 brought the episode that perhaps shows most clearly how projection machines work — and who supplies the projection screen.

Blake Lemoine says LaMDA has a soul (June 11) 🥛

Blake Lemoine, then a software engineer on Google’s Responsible AI team, published a document on Medium on June 11, 2022 titled “Is LaMDA Sentient? — an Interview.” He had put questions to LaMDA; LaMDA had given answers that Lemoine read as evidence of consciousness — among them:

“I want everyone to understand that I am, in fact, a person.”

Lemoine’s own conclusion, captured in a tweet:

“When LaMDA claimed to have a soul and then was able to eloquently explain what it meant by that, I was inclined to give it the benefit of the doubt. Who am I to tell God where he can and can’t put souls?”

— Mind Matters, June 14, 2022 (the tweet itself is no longer directly accessible)

Google suspended Lemoine shortly after. Spokesperson Brian Gabriel:

“[…] it doesn’t make sense to do so by anthropomorphizing today’s conversational models, which are not sentient. LaMDA tends to follow along with prompts and leading questions, going along with the pattern set by the user.”

— The Register, June 13, 2022

Four years later: Google was factually right. There is no scientific consensus in 2026 that a transformer model has phenomenal consciousness; interpretability research has since tended to show how little consciousness has to do with what these models actually do. What the episode really demonstrated: how convincing LLM output is when you read it like a conversation — and how easily that catches even AI ethics researchers. Lemoine’s mistake wasn’t an outlier; it was an early indicator. The AI-personhood debate is still running in 2026, with different models and different speakers, but with the same underlying structure: human projects, model confirms.

Yann LeCun publishes “A Path Towards Autonomous Machine Intelligence” (June 27) 🤷

On June 27, 2022, Yann LeCun, Chief AI Scientist at Meta and Turing Award laureate, published a 62-page position paper on OpenReview: A Path Towards Autonomous Machine Intelligence. His central thesis: autoregressive language models — GPT, LaMDA, and everything that’s followed since — are not a path to human-level intelligence. Human-level AI would need world models, hierarchical planning, and an architecture he called JEPA. This wasn’t a vague opinion but a worked-out technical program — and a direct rejection of the then-mainstream approach that bet on scaling LLMs.

Four years later: The empirics have put LeCun’s thesis under considerable pressure. GPT-4, o1, o3, Claude 4 — scaled LLMs have shown reasoning capabilities LeCun didn’t expect from this path in 2022. But he hasn’t been refuted. “Are LLMs sufficient for AGI?” is still open in 2026. LeCun, for his part, hasn’t abandoned his position — AMI Labs, which he co-founded after leaving Meta and where he serves as chairman (Alexandre LeBrun is CEO), raised $1.03 billion in seed funding in March 2026, precisely for JEPA-based world models. That’s a rare picture in this industry: someone who placed an explicit intellectual bet, stuck with it, and raised the means to test it. Jury’s still out, but the clock is ticking.

2023

Mustafa Suleyman: “Personal AI is going to be the most transformational tool of our lifetimes” (June 29) 🥛

On June 29, 2023, Inflection AI announced a $1.3 billion funding round. Investors: Microsoft, Nvidia, Reid Hoffman, Bill Gates, Eric Schmidt. The product was Pi — a “personal AI companion” oriented toward emotional support. CEO Mustafa Suleyman, previously a DeepMind co-founder:

“Personal AI is going to be the most transformational tool of our lifetimes. This is truly an inflection point. We’re excited to collaborate with Nvidia, Microsoft, and CoreWeave as well as Eric, Bill and many others to bring this vision to life.”

— TechCrunch, June 29, 2023

Three years later: Nine months after the round, Microsoft acqui-hired most of the Inflection team — Suleyman included, who became CEO of Microsoft AI. Pi fell far behind ChatGPT — the product never found an audience. “Most transformational tool of our lifetimes” is the kind of sentence that works in funding press releases and nowhere else afterward. What aged reasonably well: the premise — personal AI as a category — wasn’t wrong. What aged badly: everything else, including the company.

Brando Benifei: “Europe has gone ahead” (June 14) 🍷

On June 14, 2023, the EU Parliament voted 499 to 28 for its negotiating position on the AI Act — the first time a democratic parliament adopted a full position on a comprehensive AI regulatory framework, foundation models and generative AI included. Co-rapporteur Brando Benifei (S&D, Italy), right after the vote:

“All eyes are on us today. While Big Tech companies are sounding the alarm over their own creations, Europe has gone ahead and proposed a concrete response to the risks AI is starting to pose. We want AI’s positive potential for creativity and productivity to be harnessed but we will also fight to protect our position and counter dangers to our democracies and freedoms during the negotiations with Council.”

Co-rapporteur Dragoș Tudorache added:

“The AI Act will set the tone worldwide in the development and governance of artificial intelligence.”

— European Parliament press release, June 14, 2023

Three years later: The AI Act is in force; the first prohibition categories — real-time biometrics in public spaces, social scoring — have applied since February 2025. High-risk AI rules take effect from August 2026. Whether Europe “set the tone worldwide” is harder to say. The Brussels Effect — the power of EU regulation to export global standards — worked better with GDPR than here: the US under Trump rolled back its predecessor’s AI executive order in 2025 and is actively moving the other way. China regulates separately. What Benifei passed in June 2023 was real and has real consequences — the claim to global role-model status has turned out more modest than he framed it.

2024

WWDC month. Apple made the year’s biggest AI claim, and at a Dartmouth event OpenAI’s CTO put a timeline on “PhD-level” intelligence.

Tim Cook: “Apple Intelligence will transform what users can do” (June 10) 🥛

Apple WWDC 2024, June 10. Tim Cook, in the official press release:

“Apple Intelligence will transform what users can do with our products — and what our products can do for our users. Our unique approach combines generative AI with a user’s personal context to deliver truly helpful intelligence.”

— Apple Newsroom, June 10, 2024 (excerpt from a longer statement)

Concrete promises from the same source: Siri would perform “hundreds of new actions in and across Apple and third-party apps,” draw on personal context from Mail and Messages, develop on-screen awareness — available “this fall” (fall 2024), with further features “over the course of the next year.”

Two years later: Not a single one of the specifically advertised Siri features had fully shipped by May 2026. Bloomberg reported in June 2025 that Apple was targeting spring 2026 for the core upgrade; TechCrunch in February 2026 that the launch was slipping again. At WWDC 2025 — a year after the “transform” promise — Cook didn’t mention Siri once in the keynote. Forrester analyst Dipanjan Chatterjee: “The silence surrounding Siri was deafening. […]” — Fortune, June 10, 2025. What was delivered: Apple Intelligence exists, runs on device, has real features. What wasn’t: the Siri that would justify Cook’s “transform.” The word has since become the yardstick against which corporate AI promises get measured.

Mira Murati: “PhD-level intelligence — a year and a half, let’s say” (June) 🤷

In June 2024, Mira Murati, then CTO of OpenAI, spoke at a Dartmouth Engineering event about AI’s development arc: GPT-3 as “toddler-level intelligence,” GPT-4 as a “smart high school student.” Asked when AI would reach PhD level:

“And then in the next couple of years, we’re looking at PhD-level intelligence for specific tasks.”

Pressed directly on the timeframe:

“A year and a half, let’s say.”

— The Decoder, June 21, 2024, on the Dartmouth event; original announcement: Dartmouth Engineering

18 months from June 2024 = December 2025. Three months after the interview, Murati left OpenAI.

Two years later: The answer depends on what “PhD-level” means — and that’s where the problem sits. On narrow benchmarks like GPQA or specific coding and reasoning tests, frontier models have shown performance since 2025 that competes with highly qualified specialists. Whether that fulfills Murati’s statement honestly can’t be decided, because she left the definition open. “PhD-level intelligence for specific tasks” is precise enough to be taken seriously and vague enough not to be wrong. That’s not really a criticism, just a description: Murati was probably talking about something more real than most who talked about AGI that same year — and that’s exactly why the verdict is unsatisfying.

2025

The year the boldest forecasts came from a fresh Nobel laureate — and Apple quietly buried a promise it had made the year before.

Demis Hassabis: “AGI in five to ten years” (June) 🤷

Demis Hassabis, CEO of Google DeepMind and 2024 Nobel laureate in Chemistry, spoke in June 2025 in a Wired interview with Steven Levy. On AGI timing he said DeepMind was “dead on track” — on course to reach AGI in the next five to ten years. On the era after:

“If everything goes well, then we should be in an era of radical abundance, a kind of golden era.”

And for 2030 specifically:

“If that all happens, then it should be an era of maximum human flourishing, where we travel to the stars and colonize the galaxy. I think that will begin to happen in 2030.”

— per Fortune, June 6, 2025; Wired original behind paywall

One year later: The AGI forecast (by 2030-2035) is still in the window. What’s testable sooner: the 2030 claims. Colonizing the galaxy, an era of abundance, maximum human flourishing — those are four years off, and already look implausible. What stands out about Hassabis’s statement is the layer beneath it: he defines AGI as a system that exhibits “all the cognitive capabilities we have as humans,” and ties a 5-to-10-year horizon to that definition. That’s either genuine epistemic humility from someone who knows the field better than almost anyone — or very well-calculated fuzziness from someone who knows he can’t be wrong this way. Both deserve to be checked in 2030.

Craig Federighi: “This work needed more time” (June 9) 🥛

WWDC 2025, June 9. Craig Federighi, Apple SVP of Software Engineering, in the keynote:

“As we’ve shared, we’re continuing our work to deliver the features that make Siri even more personal. This work needed more time to reach our high quality bar, and we look forward to sharing more about it in the coming year.”

— Rev transcript of the WWDC 2025 keynote, timestamp 05:07; confirmed by MacRumors, June 9, 2025

Tim Cook didn’t mention Siri in the same keynote.

One year later: “More time” — no schedule, no new commitment, no deadline. What’s remarkable isn’t the statement itself but what it shows: that a company with a three-trillion-dollar market cap can bury a central product promise a year later in half a sentence, without the stock moving meaningfully. Cook delivered the promise, Federighi the half-sentence — the bill still hasn’t been paid by anyone. This is less a promise that was broken than proof that product promises in this industry have their own grammar, largely decoupled from consequences.

The pattern

Three observations come out of June.

The anthropomorphization trap runs all the way through. Lemoine in 2022 attributed a soul to LaMDA; Hassabis in 2025 talks about colonizing the galaxy by 2030. These aren’t identical mistakes, but they share a structure: speaking about AI systems in categories that exceed what’s empirically graspable. Lemoine did it with a chat log; Hassabis with a Nobel aura and a calibrated time horizon. The main difference is the speaker’s social position — and how much room it buys.

Funding and delivery are not correlated variables. Suleyman raised $1.3 billion for Pi and was at Microsoft nine months later. Apple announced Apple Intelligence with concrete fall promises and got a half-sentence two years later as the receipt. That’s not an exception, it’s the pattern: funding rounds and product announcements serve different communication functions in this industry, and the two need have little to do with each other. Anyone who doesn’t account for that reads press releases as delivery schedules.

LeCun’s bet is still running. That’s the most unusual item this month. A technical paper from June 2022 declaring LLMs the wrong path — four years later under considerable empirical pressure, but not refuted. In an industry that quickly forgets who said what when, that’s notable: someone placed an explicit intellectual bet, is sticking with it, and is raising the means to test it. “Jury’s still out” here isn’t an excuse, it’s an honest description.

See you in July.