About
One person reading narrow slices of the AI literature in public, with nothing to sell and no result to protect.
150 × 188
I read narrow slices of the AI research literature closely and write up where I think each one has got to, in language a technical reader outside the field can act on.
[YOUR BACKGROUND — two or three sentences. What you have worked on, for how long, and what gives you the standing to read this material. Be specific and factual rather than impressive. If your background is applied rather than academic, say so plainly: someone who has had to make research work in production is often a better judge of practical viability than a specialist, and readers know it.]
[OPTIONAL — one sentence on what led you to start doing this publicly.]
Why this is not a commercial site
Independence is the only thing this work has going for it. A survey is worth reading precisely because the person writing it has no stake in the answer, and that is difficult to maintain once there is a client, a sponsor, or a product in the picture.
So there is nothing to buy here, no advertising, no sponsorship, and no consulting funnel behind the analysis. The site is run as a personal, non-commercial project.
What I don’t do
I do not review work on request, evaluate proposals privately, endorse products, or act as a reference in disputes between researchers. I choose domains because they interest me, or because the claims being made around them seem to be outrunning the evidence.
This is not peer review and carries none of its authority. It is one person reading carefully in public, and it should be weighed accordingly.
How the work is done
Every finding follows the same process: fix a domain boundary, collect what falls inside it, group the proposals by the mechanism they actually use, judge which of them look usable outside a benchmark, and occasionally run the code. The method page sets that out in full, including what I record and where my judgement enters.
Getting in touch
Write to contact@benchcraft.org.
I am glad to hear from authors whose work I have covered, particularly if you think I have misread something. Corrections are published alongside the original rather than folded in quietly — see the disclaimer for how that works.
I do not accept guest posts, link exchanges, or promotional material.