Reported model capabilities rest on benchmark scores that need careful interpretation.
What benchmarks are
Fixed sets of tasks with known answers.
Which allow comparison between systems.
Contamination
Test items appearing in training data.
Which inflates scores and is hard to rule out.
Saturation
Benchmarks becoming uninformative once scores cluster near the top.
Real-world correlation
Benchmark performance predicting usefulness imperfectly.
Which practitioners consistently observe.
Why benchmark scores mislead
The test items may have appeared in training data, the tasks may not resemble what users actually do, and the scores are reported by the organisation with an interest in them being high.
Which is not an accusation of dishonesty so much as a description of the incentives.
Independent evaluation exists and is under-resourced relative to the scale of the claims being made.
Contamination detection
Methods for identifying memorised test items.
Which are imperfect.
Held-out and dynamic benchmarks
Test sets kept private or regenerated.
Which addresses contamination partially.
Human evaluation
People comparing outputs directly.
Which is expensive and more informative.
What to look for in claims
Which benchmark, which version, and who ran the evaluation.
Why the mechanics are worth understanding
These systems are described in language that suggests understanding, reasoning and knowledge, and the underlying operations are considerably more specific than those words imply. That gap is where most confusion, most disappointment and a substantial amount of misplaced trust originates.
Someone who knows that a language model predicts tokens, that it has no separate lookup step, that its context is a hard limit and that its confidence is unrelated to its accuracy will make better decisions about when to use one than someone working from the marketing description. Not because the technology is worse than claimed, but because knowing what a tool actually does is how anyone uses a tool well.
The pattern across all of this
Nearly every behaviour that people find surprising follows directly from a design choice that is documented and comprehensible. Invented citations follow from the training objective. Difficulty with spelling follows from tokenisation. Degradation in long conversations follows from context limits and attention behaviour. Cost structures follow from the fact that inference is not free.
None of these are mysteries or failures of implementation. They are consequences, and they are predictable once the mechanism is understood.
Where the reliable information is
Research papers, model documentation, technical blogs from the organisations building these systems, and independent evaluation work. Most of it is freely available and considerably more measured than either the promotional material or the criticism that circulates.
The field moves quickly enough that any general article, including this one, dates rapidly. Anything with consequences attached is worth checking against current documentation rather than against a description written at some point in the past.
What follows from all this in practice
Use these systems for work where the output can be checked, where fluency is genuinely useful, and where a draft that needs revision is more valuable than a blank page. Avoid relying on them for specific facts, citations, numbers or anything where a confident wrong answer is worse than no answer.
That is a narrower recommendation than the enthusiasm suggests and a much broader one than the dismissal allows. Both positions tend to be argued by people who have not looked closely at what the systems do, and the useful territory between them is large.
On the pace of change
Capabilities have moved quickly enough that specific claims about what models can and cannot do have a short shelf life. Several limitations described confidently a few years ago have been substantially reduced, and several that were expected to fall have not.
The underlying mechanisms have changed far less than the capabilities. Next token prediction, tokenisation, context limits and the absence of a verification step have all remained, which is why the same categories of failure keep recurring in new forms even as performance improves.
A note on evaluating claims
When a capability is announced, useful questions are: measured on what benchmark, compared against what, evaluated by whom, and reproducible by anyone outside the organisation making the claim. Those four questions dispose of a considerable amount of what circulates.
Independent replication in this field is under-resourced relative to the scale of the claims, which is a structural problem rather than a criticism of any particular organisation.
Further reading
Technical documentation from model providers, peer-reviewed research, and independent evaluation organisations are the three places where this material is covered accurately.
All three are freely accessible, and all three are considerably drier and more useful than the commentary built on top of them.
One last observation
The most useful thing anyone can know about these systems is what they are optimising for, because everything else follows from it. A model trained to produce plausible text will produce plausible text, reliably and regardless of whether the text is true.
That is not a criticism. It is a specification, and reading it as one makes the whole field considerably easier to think about.