# Hassabis hasn't moved his AGI forecast in 16 years — the news is the drug pipeline

URL: https://www.thedeepfeed.ai/posts/2026-05-01-hassabis-2030-agi-disease-compression/
Category: Research
Published: 2026-05-07
Author: the-deep-feed
Tags: hassabis, deepmind, isomorphic-labs, alphafold, agi-timeline, sequoia
Kind: deep

> Hassabis's "2030" line at AI Ascent 2026 is the same prediction he made in 2010. The publishable claim is the falsifiable one: AlphaFold 3 plus Isomorphic Labs has now slipped its first clinical trial to end of 2026.

## TL;DR

- Sequoia titled it *"Three Quarters of the Way to AGI"* — fraction never appears in the transcript. Hassabis says **"2030. I've been pretty consistent."** Same date as his **2010** "20-year mission."
- Verifiable bet isn't AGI — it's **Isomorphic Labs**. $600M from Thrive (Mar 2025), $2.9B with **Lilly** + **Novartis**, first-in-human trial just slipped from 2025 to **end of 2026**.
- Compression claim has a number: drug discovery from **a decade and a billion dollars** to **months, weeks, perhaps days**. He dates AGI; hedges bio as "the next few years." Asymmetry is the story.
- **AGI-2030 is now median**, not bullish. Altman: researcher-class AI by **March 2028**. Amodei: powerful AI 2026–2027. **Yann LeCun** calls it all "a dead end" and left Meta in January 2026.
- Karpathy's **"decade of agents"** (Part 1) and Hassabis's **AGI by 2030** can't both be right about the same five years. The disagreement, not the consensus, is what founders should plan against.

## The forecast that didn't move

The YouTube cut is titled ["We're Three Quarters of the Way to AGI"](https://youtu.be/AFpeWo1GTeg). The phrase appears nowhere in the 5,329-word transcript. **Demis Hassabis** does not give a percentage, a benchmark, or a metric for it. What he does, sitting across from Sequoia partner Pat Grady at AI Ascent 2026, is something more interesting and far more verifiable: he restates a prediction he made in 2010 and asserts it is on track.

![A clock face with the hour hand fixed at 2030 and three layered planes below marking 2010, 2018, 2026: the forecast that didn't move](/post-images/2026-05-01-hassabis-2030-agi-disease-compression/hero-2030-clock.jpg)

Asked "over/under on distribution, year of AGI," he answers without preamble:

> "2030. I've been pretty consistent about that."

He has been. The earliest archived version of the line is from a DeepMind founding-era talk preserved in [a quote archive](https://ceointerviews.ai/quote/when-we-started-deepmind-back-in-2010-we-OTQ4MzUw/): *"When we started DeepMind back in 2010 we thought this would be a 20-year kind of mission to build AGI."* Twenty years from 2010 is 2030. He restated it [in 2019](https://www.deepmind.com/blog/announcements/entering-our-tenth-year-at-deepmind), to [Lex Fridman in July 2025](https://lexfridman.com/demis-hassabis-2-transcript/) ("50% chance by 2030"), to [60 Minutes in April 2025](https://www.cbsnews.com/news/artificial-intelligence-google-deepmind-ceo-demis-hassabis-60-minutes-transcript/) ("five to 10 years away"), [in TIME's 2025 cover profile](https://time.com/7277608/demis-hassabis-interview-time100-2025/), and [in December 2025 to Axios](https://www.axios.com/2025/12/05/ai-deepmind-gemini-agi). The Sequoia talk is the latest restatement, not a new one.

In the talk he is even more explicit about the through-line:

> "It's gone, you know, on the absolutely amazing side of the optimistic side of what we thought. Still actually within what we were predicting in 2010. We thought it would be a 20-year mission. And I think we're basically exactly on track as a field."

That is the headline most reporting buried. A sitting Google DeepMind CEO is not making a five-year prediction in 2026; he is checking off **year 16 of a 20-year prediction made in 2010**, the year the company was founded. Whether you find that compelling or self-flattering depends on whether you grade him on the date or on what's actually been built between the two.

## A 16-year forecast, on the same date

![Eight irregular timeline markers, rightmost in red — sixteen years of Hassabis restating the same date](/post-images/2026-05-01-hassabis-2030-agi-disease-compression/hassabis-timeline-stack.jpg)

The consistency is unusual enough to tabulate. The dates and quotes below come from primary sources where available; the *"five to 10 years"* phrasing recurs because Hassabis treats AGI as a probability distribution centered on 2030, not a point estimate.

| Year | Venue | Position | Source |
| --- | --- | --- | --- |
| 2010 | DeepMind founding | "20-year mission to build AGI" → ~2030 | [Quote archive](https://ceointerviews.ai/quote/when-we-started-deepmind-back-in-2010-we-OTQ4MzUw/) |
| 2019 | DeepMind 10-year retrospective | "Working on what could be one of the greatest experiments in history" — same mission framing | [DeepMind blog](https://www.deepmind.com/blog/announcements/entering-our-tenth-year-at-deepmind) |
| Feb 2024 | Hard Fork podcast | AGI-grade systems "5–10 years away" | [Hard Fork EP 71](https://www.youtube.com/watch?v=nwUARJeeplA) |
| Apr 2025 | 60 Minutes | "Five to 10 years away" — i.e., 2030–2035 | [CBS News](https://www.cbsnews.com/news/artificial-intelligence-google-deepmind-ceo-demis-hassabis-60-minutes-transcript/) |
| Apr 2025 | TIME100 cover | "Preparing for AI's endgame"; 2030 cited explicitly | [TIME](https://time.com/7277608/demis-hassabis-interview-time100-2025/) |
| Jul 2025 | Lex Fridman #475 | "50% chance by 2030" | [Lex Clips](https://www.youtube.com/watch?v=T9Knc3Mdcec) |
| Dec 2025 | Axios AI+ Summit | Category-defining AGI on the horizon, same window | [Axios](https://www.axios.com/2025/12/05/ai-deepmind-gemini-agi) |
| Apr 2026 | Sequoia AI Ascent | "2030. I've been pretty consistent about that" | [YouTube](https://youtu.be/AFpeWo1GTeg) |

This is what a forecast looks like when it doesn't have to be revised. It is the opposite of the OpenAI pattern, where Sam Altman has [moved from "AGI in 2025"](https://www.ainews.com/p/sam-altman-predicts-agi-by-2025-in-openai-s-bold-vision-for-ai) to ["research interns in September 2026, AGI-class researchers by March 2028"](https://www.resultsense.com/news/2025-11-01-openai-roadmap-ai-research-interns-by-2026-agi-researchers-by-2028) inside a single calendar year.

The framing matters because it changes what the news is. The news isn't "Hassabis predicts AGI by 2030." The news is that the field has caught up to the median Hassabis prediction without him having to move the goalposts.

## Why "three quarters" is the wrong frame

![Six forecast points at varying heights, leftmost in red — a spread, not a consensus on AGI timelines](/post-images/2026-05-01-hassabis-2030-agi-disease-compression/forecast-spread.jpg)

The Sequoia title is a crowd-pleaser; it isn't supported by anything Hassabis says on stage. He never claims a fraction. When he is pressed on what the gap actually is, he points at well-known holes:

> "We're in the midst of the agent era now. But then there's a further step of like, you know, does it have agency? Is it conscious?"

[**Andrej Karpathy**, the day before](/posts/2026-05-01-karpathy-software-3-agentic-engineering/), framed those same gaps explicitly as a **decade** of work — coding agents that one-shot tasks, but jagged intelligence everywhere else, and verifiability as the master variable. **Hassabis** and **Karpathy** sat on the same Sequoia stage 24 hours apart and gave time-on-target estimates that differ by a factor of two. Both numbers cannot be right about the same five years.

That disagreement, with no diplomatic hedge between them, is the publishable thing about AI Ascent 2026. The room of LPs and founders heard *"AGI by 2030"* on Tuesday and *"decade of agents"* on Wednesday and was supposed to leave with conviction.

The honest comparison across the field looks like this:

| Person | Org | Public timeline | Definition used | Source |
| --- | --- | --- | --- | --- |
| **Hassabis** | Google DeepMind | **2030** (~50% probability) | "All cognitive capabilities humans have" | [Sequoia 2026](https://youtu.be/AFpeWo1GTeg), [Lex 2025](https://lexfridman.com/demis-hassabis-2-transcript/) |
| **Altman** | OpenAI | "Research interns" Sep 2026; researcher-class **Mar 2028** | OpenAI's 5-level framework | [OpenAI roadmap](https://www.resultsense.com/news/2025-11-01-openai-roadmap-ai-research-interns-by-2026-agi-researchers-by-2028) |
| **Amodei** | Anthropic | "Powerful AI" by **2026–2027**, civilizational impact through 2030 | "Country of geniuses in a datacenter" | [Machines of Loving Grace](https://www.darioamodei.com/essay/machines-of-loving-grace) |
| **Karpathy** | independent | **Decade of agents**, not a year | Agentic engineering raises ceiling slowly | [Part 1, TDF](/posts/2026-05-01-karpathy-software-3-agentic-engineering/) |
| **LeCun** | post-Meta lab | "Years away," LLMs are a dead end | Requires world-model architecture | [Forbes Davos 2026](https://www.youtube.com/watch?v=5PQtJxd4U0M) |
| **Marcus** | NYU / commentator | Not in this decade on current architectures | Symbolic + neural needed | [Marcus on AI](https://garymarcus.substack.com/p/the-last-few-months-have-been-devastating) |

There is a four-way split: Hassabis and Altman near the same date but with very different definitions; Amodei more aggressive on capability than year; Karpathy halving Hassabis's slope; LeCun and Marcus rejecting the framing entirely. The "consensus" in trade press coverage is a confidence interval that spans **three years to multiple decades**, which is the same as no consensus at all.

## The other claim — and why it's the better one

![An axonometric funnel collapsing through five layered planes into a red beaker — drug discovery compression, decade to days](/post-images/2026-05-01-hassabis-2030-agi-disease-compression/compression-funnel.jpg)

Hassabis spends more of the talk on biology than on AGI dates. The drug discovery argument is the part that is concrete enough to falsify. His version, on stage at Sequoia:

> "Instead of taking like, you know, an average of 10 years, drug discovery times — down to months, maybe even weeks, perhaps even days one day. And then I think then all disease could be in reach."

This is a specific, compressible timeline claim from the CEO of [**Isomorphic Labs**](https://www.isomorphiclabs.com/), which spun out of DeepMind in November 2021 with the explicit mission to do exactly this. Unlike "AGI by 2030," it is grounded in a stack of artifacts, contracts, and milestones a journalist can actually check.

The artifacts are real. **AlphaFold 2** mapped 200M+ proteins. **AlphaFold 3**, [published in *Nature* in May 2024](https://www.nature.com/articles/s41586-024-07487-w), generalized the model from proteins-only to "all life's molecules": [proteins, DNA, RNA, ligands, ions and modified residues](https://deepmind.google/blog/alphafold-3-predicts-the-structure-and-interactions-of-all-lifes-molecules/), with reported accuracy gains of **at least 50%** on protein-ligand interactions versus the best classical methods. It is the model on which Hassabis and **John Jumper** [shared the 2024 Nobel Prize in Chemistry](https://www.nobelprize.org/prizes/chemistry/2024/) with **David Baker**.

The contracts are real. In January 2024, Isomorphic announced [its first pharma deals](https://endpts.com/alphabets-ai-unit-isomorphic-inks-drug-discovery-deals-with-eli-lilly-novartis-for-up-to-3b/) — **$45M upfront** from **Eli Lilly** with up to **$1.7B in milestones**, plus **$37.5M upfront** from **Novartis** with up to **$1.2B more**, for a [combined headline of nearly $3B](https://www.pharmamanufacturing.com/development/process-development/news/33017419/isomorphic-labs-signs-ai-drug-discovery-deals-with-novartis-lilly). In March 2025, the company [closed its first external round](https://www.isomorphiclabs.com/articles/isomorphic-labs-announces-600m-external-investment-round): **$600M led by Thrive Capital**, with **GV** and Alphabet participating, earmarked for "advancing therapeutic programs into the clinic."

The milestones are real, and one of them just slipped. In July 2025, [Fortune reported](https://fortune.com/2025/07/06/deepmind-isomorphic-labs-cure-all-diseases-ai-now-first-human-trials/) that Isomorphic was "preparing to dose the first patients" — the first humans to receive a fully AI-designed drug from this pipeline. In January 2026, [Reuters reported](https://www.reuters.com/business/healthcare-pharmaceuticals/google-backed-ai-drug-discovery-startup-isomorphic-labs-delays-clinical-trial-2026-01-20/) Hassabis at Davos pushing that timeline to "end of 2026," a roughly 12-month delay. Then in February 2026, Isomorphic president **Max Jaderberg** [told Endpoints News](https://endpoints.news/isomorphic-claims-major-advance-with-new-ai-drug-design-engine) that the team had built a "step change" successor to AlphaFold 3 internally.

That is what a falsifiable infrastructure claim looks like under test.

## The Isomorphic milestone ledger

| Milestone | Date | Detail | Source |
| --- | --- | --- | --- |
| Spinout from DeepMind | Nov 2021 | "AI-first drug discovery" | [Isomorphic launch post](https://www.isomorphiclabs.com/articles/introducing-isomorphic-labs) |
| AlphaFold 3 paper | May 2024 | Generalizes structure prediction to DNA/RNA/ligands | [*Nature* 2024](https://www.nature.com/articles/s41586-024-07487-w) |
| Eli Lilly + Novartis deals | Jan 2024 | $83M upfront, **$2.9B** in milestones | [Endpoints](https://endpts.com/alphabets-ai-unit-isomorphic-inks-drug-discovery-deals-with-eli-lilly-novartis-for-up-to-3b/) |
| Nobel Prize in Chemistry | Oct 2024 | Hassabis + Jumper share half of prize | [Nobel](https://www.nobelprize.org/prizes/chemistry/2024/) |
| $600M Series A | Mar 2025 | Thrive Capital lead; first external round | [Isomorphic press](https://www.isomorphiclabs.com/articles/isomorphic-labs-announces-600m-external-investment-round) |
| First human trial planned | Jul 2025 (target: 2025) | Oncology candidate first | [Fortune](https://fortune.com/2025/07/06/deepmind-isomorphic-labs-cure-all-diseases-ai-now-first-human-trials/) |
| Clinical trial slipped | Jan 2026 | New target: end of 2026 | [Reuters](https://www.reuters.com/business/healthcare-pharmaceuticals/google-backed-ai-drug-discovery-startup-isomorphic-labs-delays-clinical-trial-2026-01-20/) |
| Next-gen design engine | Feb 2026 | Successor to AlphaFold 3 internally | [Endpoints](https://endpoints.news/isomorphic-claims-major-advance-with-new-ai-drug-design-engine) |

The slip is the most newsworthy line in that table. *"Decade to days"* compresses well in a fireside chat; the wet-lab stack is still slow enough that a single AI-designed oncology candidate took **roughly two years** from AlphaFold 3's release to the first dosing window — and missed it. That isn't a refutation of the thesis. It's the first piece of evidence about the actual slope.

## "ML is the description language for biology" — what the claim is

Underneath the dates there is a stronger epistemological claim, the one that explains the company structure. Hassabis on stage:

> "Machine learning is the perfect description language for biology in the same way mathematics is for physics."

And:

> "In biology and in lots of these natural systems you have loads of weak signals, weak correlations, tons of data far too much that any human mind can analyze. So machine learning is the perfect tool to describe those kinds of systems where until today mathematics hasn't been able to do that."

Read literally, this is a claim about *expressive power*, not productivity. Physics yielded to closed-form math because its domain has clean conservation laws and a small number of relevant variables per problem. Biology, with proteins folding under quantum-mechanical bond effects, cells as dynamical emergent systems, and drug binding as a probabilistic surface-fit problem, does not. Hassabis's bet is that learned models can be the *formal substrate* for these systems the way differential equations were for electromagnetism.

If the bet is right, AlphaFold 3 is not a one-off; it is the first instance of a class. DeepMind has already shipped [GraphCast / WeatherNext](https://deepmind.google/blog/10-years-of-alphago/), the most accurate medium-range global forecasting system, as another instance — the same machinery aimed at a fluid-dynamical emergent system. Hassabis explicitly cites "virtual cell" simulators as the next target on stage:

> "We're working on a kind of what I call a virtual cell. So you know hugely dynamical emergent system."

This is the Karpathy bet from the other side of the building: **Karpathy** [calls neural networks the new programming surface — Software 3.0](/posts/2026-05-01-karpathy-software-3-agentic-engineering/). **Hassabis** calls them the new modeling surface for systems formal math could never reach. Both descriptions can be true, but they imply different startup categories. Software 3.0 implies the moat is the verifier and the data. Description-language-for-biology implies the moat is **the wet-lab loop and the regulatory pathway**, which is exactly what the $600M raise and the Lilly/Novartis deals are designed to build.

## The opposing camp is also organizing

[**Yann LeCun**](https://www.youtube.com/watch?v=5PQtJxd4U0M), at Davos in January 2026, called LLMs a dead end and human-level AI "years away," then [left Meta the same month](https://creati.ai/ai-news/2026-01-26/yann-lecun-warns-ai-industry-wrong-path-departs-meta/) to build something else. **Gary Marcus** [in October 2025 declared "game over" on the LLM-to-AGI thesis](https://garymarcus.substack.com/p/the-last-few-months-have-been-devastating); [in December 2025](https://garymarcus.substack.com/p/breaking-news-scale-is-all-you-need) he claimed NeurIPS itself had vindicated his critique that scale alone is not enough. Both [are quick to point out that Hassabis is now also pushing world models](https://garymarcus.substack.com/p/further-breaking-news-further-vindicating) — which Marcus reads as an implicit concession that pure-LLM scaling won't get to AGI on its own.

Hassabis's position in the talk is consistent with that read. He calls the agent era a step "in the midst of," not the destination. He cites simulations, world models, virtual cells, and learned simulators for economics as the *next* science, distinct from the autoregressive sequence-prediction paradigm:

> "Learning simulators basically would it be — these are in domains where we don't know the mathematics of it well enough or it's perhaps too complex. We can't just write down a special case simulator."

The interesting thing is that the LeCun / Marcus camp and the Hassabis camp **agree on the architecture story** (pure LLM scaling is necessary but not sufficient) and **disagree only on the year**. LeCun says "years"; Marcus says "not in this decade"; Hassabis says "2030." The actual delta is whether grafting reinforcement learning, tool use, self-play, and learned simulators onto LLM substrates closes the gap inside five years or not. That is an engineering question, not a metaphysical one, and it is testable.

## Stop reading the AGI year. Start reading the trial readout.

![A folded protein on the left, three charcoal arrows pointing right, a single red pill on the right: from AlphaFold to a dosed candidate](/post-images/2026-05-01-hassabis-2030-agi-disease-compression/fold-to-cure.jpg)

The more an AGI prediction becomes a date on a calendar, the less information it carries. Hassabis is the rare practitioner whose date hasn't drifted, but the date is also the least valuable part of the talk for anyone making a decision. Five-year AGI prediction asymmetry has become a trade press subgenre; founders building right now can't act on it.

What founders can act on is the [**$3B in milestone payments**](https://endpts.com/alphabets-ai-unit-isomorphic-inks-drug-discovery-deals-with-eli-lilly-novartis-for-up-to-3b/) sitting against an AlphaFold 3 stack, the [**$600M Thrive round**](https://www.isomorphiclabs.com/articles/isomorphic-labs-announces-600m-external-investment-round) earmarked for moving programs into the clinic, the [**slipped 2026 trial**](https://www.reuters.com/business/healthcare-pharmaceuticals/google-backed-ai-drug-discovery-startup-isomorphic-labs-delays-clinical-trial-2026-01-20/), and the [**successor model**](https://endpoints.news/isomorphic-claims-major-advance-with-new-ai-drug-design-engine) the Isomorphic team is already trying to publish past AlphaFold 3.

That is also the cleanest test of the bigger thesis. If "ML is to biology what math is to physics" is right, the next 36 months should look like this: the first AI-designed candidate dosed in humans before December 2026; at least one Lilly or Novartis milestone triggered by 2027; and a credible second-generation drug-design engine in production at Isomorphic. If those happen, "all disease in reach this century" stops being a slogan and starts being a base case.

If they don't — if the trial slips again, if no milestone triggers, if the next-gen engine ships without measurable gains, the more honest reading is that the AGI-to-medicine pipeline is bottlenecked by exactly the things AI was supposed to solve: wet-lab biology, regulators, and the fact that the human body is harder than the protein database.

Hassabis already gave the right framing for which bet matters more. On the same stage, asked for his proudest moment at DeepMind, he didn't say "Gemini" or "Project Astra" or anything tied to the agent era. He said one word:

> "AlphaFold."

The rest of the field is still arguing about the AGI year. He's already moved the conversation to the molecule. The bet The Deep Feed is making is that the molecule is the part to watch — because it's the only part of his talk where the deadline can actually be checked.

For the other half of this disagreement (Karpathy's case that the next decade belongs to agentic engineering, not AGI) see [Part 1 of this series](/posts/2026-05-01-karpathy-software-3-agentic-engineering/). For why the same Sequoia event is also a thesis on services-as-software economics, see [the services-as-software post](/posts/2026-04-30-sequoia-services-as-software-thesis/).

## Sources

- [Sequoia Capital — Demis Hassabis: We're Three Quarters of the Way to AGI (AI Ascent 2026)](https://youtu.be/AFpeWo1GTeg)
- [CEO Interviews archive — Hassabis: "20-year mission" quote (DeepMind founding)](https://ceointerviews.ai/quote/when-we-started-deepmind-back-in-2010-we-OTQ4MzUw/)
- [Google DeepMind — Entering our tenth year at DeepMind (Hassabis, Dec 2019)](https://www.deepmind.com/blog/announcements/entering-our-tenth-year-at-deepmind)
- [Google DeepMind — AlphaGo at 10: from games to biology and beyond (Hassabis, Mar 2026)](https://deepmind.google/discover/blog/alphago-at-10-how-ai-innovation-is-paving-the-path-to-agi/)
- [60 Minutes — AI could end disease, lead to "radical abundance" (Apr 2025 / updated 2025)](https://www.cbsnews.com/news/artificial-intelligence-google-deepmind-ceo-demis-hassabis-60-minutes-transcript/)
- [TIME100 — Demis Hassabis Is Preparing for AI's Endgame (Apr 2025)](https://time.com/7277608/demis-hassabis-interview-time100-2025/)
- [Lex Fridman Podcast #475 — Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games (Jul 2025)](https://lexfridman.com/demis-hassabis-2-transcript/)
- [Lex Clips — "Timeline for AGI: 2030 with 50% chance" (Hassabis, Jul 2025)](https://www.youtube.com/watch?v=T9Knc3Mdcec)
- [Axios — "Transformative" AGI is on the horizon, DeepMind's Hassabis says (Dec 2025)](https://www.axios.com/2025/12/05/ai-deepmind-gemini-agi)
- [Nature — Accurate structure prediction of biomolecular interactions with AlphaFold 3 (Abramson et al., 2024)](https://www.nature.com/articles/s41586-024-07487-w)
- [Google DeepMind — Introducing AlphaFold 3 (May 2024)](https://deepmind.google/blog/alphafold-3-predicts-the-structure-and-interactions-of-all-lifes-molecules/)
- [NobelPrize.org — Chemistry 2024: Baker, Hassabis, Jumper](https://www.nobelprize.org/prizes/chemistry/2024/)
- [Endpoints News — Isomorphic's first pharma deals: $83M upfront, $2.9B in milestones (Jan 2024)](https://endpts.com/alphabets-ai-unit-isomorphic-inks-drug-discovery-deals-with-eli-lilly-novartis-for-up-to-3b/)
- [Isomorphic Labs — $600M external investment round (Mar 2025)](https://www.isomorphiclabs.com/articles/isomorphic-labs-announces-600m-external-investment-round)
- [Fortune — Isomorphic gearing up for its first human trials (Jul 2025)](https://fortune.com/2025/07/06/deepmind-isomorphic-labs-cure-all-diseases-ai-now-first-human-trials/)
- [Reuters — Google-backed Isomorphic Labs delays clinical trial timeline (Jan 2026)](https://www.reuters.com/business/healthcare-pharmaceuticals/google-backed-ai-drug-discovery-startup-isomorphic-labs-delays-clinical-trial-2026-01-20/)
- [Endpoints News — Isomorphic claims "step change" versus AlphaFold 3 (Feb 2026)](https://endpoints.news/isomorphic-claims-major-advance-with-new-ai-drug-design-engine)
- [Dario Amodei — Machines of Loving Grace (Oct 2024)](https://www.darioamodei.com/essay/machines-of-loving-grace)
- [Sam Altman — OpenAI roadmap: research interns 2026, researchers 2028](https://www.resultsense.com/news/2025-11-01-openai-roadmap-ai-research-interns-by-2026-agi-researchers-by-2028)
- [Imagination In Action / Forbes — Yann LeCun: LLM era is ending (Davos, Jan 2026)](https://www.youtube.com/watch?v=5PQtJxd4U0M)
- [Marcus on AI — Game over. AGI is not imminent (Oct 2025)](https://garymarcus.substack.com/p/the-last-few-months-have-been-devastating)
- [Marcus on AI — "Scale Is All You Need" is dead (Dec 2025)](https://garymarcus.substack.com/p/breaking-news-scale-is-all-you-need)
- [Pharma Manufacturing — Isomorphic signs Lilly, Novartis ($3B) (Jan 2024)](https://www.pharmamanufacturing.com/development/process-development/news/33017419/isomorphic-labs-signs-ai-drug-discovery-deals-with-novartis-lilly)
- [Big Technology — Hassabis: The Path To AGI, LLM Creativity (interview)](https://www.bigtechnology.com/p/google-deepmind-ceo-demis-hassabis)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-05-01-hassabis-2030-agi-disease-compression/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt