“Meta’s woes with Llama 4 further confirm my March 2024 predictions that getting to a GPT-5 level would be hard, and that we would wind up with many companies with similar models, and essentially no moat, along with a price war, with profits modest at best. That is indeed exactly where we are.”
Our reading:
Meta's woes with Llama 4 further confirm Marcus's March 2024 predictions that reaching GPT-5 level would be hard and that many companies would end up with similar models, essentially no moat, a price war, and profits modest at best, and that is exactly the current situation.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
On January 22, 2025, DeepSeek reported reasoning performance comparable to OpenAI-o1-1217 [e2]. On January 31, 2025, OpenAI reported that per-token pricing had fallen 95% since GPT-4 launched [e3]. These reports support similarity on selected tasks and falling prices, but do not establish that competitive pressure caused the cuts [e2, e3]. On April 5, 2025, Meta reported benchmark improvements and said Behemoth was still training [e1]. That announcement does not establish the difficulty of reaching an unspecified GPT-5 capability level [e1]. These items do not settle the combined claim about development difficulty, competition and profits [e1, e2, e3].
Caveats
- The model comparisons are developer reports covering selected capabilities.
- The packet does not establish the presence or absence of durable competitive advantages, or a price war.
- No comparable realized net-profit evidence covers calendar 2024 for a defined provider population and accounting basis.
- GPT-5 level and modest profits have no specified thresholds.
- The original March 2024 prediction's wording and date remain unverified.
Evidence (3 items)
Context only: Meta,
still training
Why it matters: Meta's largest announced model remained unfinished, but the announcement also reports improvements and competitive benchmark results. This qualifies the characterization of the release as straightforward confirmation of industry stagnation.
- Publisher
- Meta
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
Context only: DeepSeek-AI, arXiv,
performance comparable to OpenAI-o1-1217
Why it matters: This supports similarity between competing providers on evaluated reasoning tasks. It does not establish equivalence across all capabilities or the absence of commercial advantages.
- Where
- Abstract
- Publisher
- DeepSeek-AI, arXiv
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
Context only: OpenAI,
keeping costs low
Why it matters: The surrounding passage reports a 95% reduction in per-token pricing since GPT-4 launched. This supports falling prices, but does not establish competitive causation, the absence of a moat or realized profits.
- Where
- What's next
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Conditions
- essentially no moat (their words)
- profits modest at best (their words)
Where we had to interpret
- Assess the asserted current situation as of April 6, 2025, rather than treating the retrospective reference to predictions as a new forecast.
- Read similar models as broadly competitive capabilities, and a price war as substantial competitive price cutting.
- Read no moat as little durable competitive protection for frontier model providers; similarity on selected benchmarks alone does not establish this.
- GPT-5 level ordinarily suggests a substantial advance beyond the preceding frontier, but Marcus supplies no capability threshold. A missing product name alone cannot establish difficulty.
- Use early-2025 evidence for the current technical and competitive situation. For profits, use the latest completed annual reporting period available before the post, calendar 2024; a single quarter is not consistently available across the unspecified providers.
- The overall target interval encompasses these distinct reporting windows. The financial accounting basis remains unknown because the claim supplies none.
Right if: Evidence establishes the combined situation: difficult progress toward a substantial frontier advance, broadly comparable models across providers, little durable competitive protection, competitive price reductions, and at most modest profits on a consistent financial basis.
Wrong if: Comparable evidence establishes a material contrary condition, such as a durable model-provider advantage or substantial profits across the relevant providers. A single benchmark win or a parent company's unrelated profits is insufficient.
We wrote this test after the outcome was known.
What would settle it
Date-verifiable text of the March 2024 prediction; a defined capability threshold and evidence of development difficulty by April 6, 2025; comparable early-2025 model evaluations, evidence of durable competitive advantages and competitive price cutting; and realized calendar-2024 net profits for a defined provider population, accounting basis and modest-profit threshold.
We found 12 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Scaling is over, the bubble may be deflating, LLMs still can’t reason, and you can’t trust Sam, paragraph 4.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2025-04-06.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Scaling is over, the bubble may be deflating, LLMs still can’t reason, and you can’t trust Sam
Meta’s woes with Llama 4 further confirm my March 2024 predictions that getting to a GPT-5 level would be hard, and that we would wind up with many companies with similar models, and essentially no moat, along with a price war, with profits modest at best. That is indeed exactly where we are.
- Where
- paragraph 4
- Published
- 2025-04-06
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
The final check failed: the write-up said more than the evidence showed.