Different fields have different terms for magic. In complex systems it’s emergence. In AI it’s recursive self-improvement.
RESEARCH
-
Randomness in AI evaluation makes reproducibility challenging
By
–
That is surprising, but there is a lot of randomness. If you try again from scratch, you may get better (or worse) results. It makes evaluation really hard.
-
Defining LLM Hallucinations: Terminology and Correct Definition
By
–
Quick terminology check: what would you say is closest to the correct definition of "hallucination" when talking about LLMs?
-
GPT-5 Performance: Real-World Task Improvements Over Competitors
By
–
On the big picture: GPT-5 as a model is pretty much on the same curve as the other top labs. I'd expect the usual leapfrogging between Gemini, Claude, OpenAI, & Grok to continue. Where there are some big gains is that GPT-5 seems well-trained for real world tasks in new ways.
-
PhD holders unlikely to produce flawed data visualizations
By
–
if they have a phd then there’s no way they would’ve made this graph after two phds
-

Superintelligence Claims and Diminishing Returns Logic
By
–
Superintelligence is near, provided you convince yourself that diminishing returns are increasing ones.
-

PhD-level rigor in data visualization and research standards
By
–
people arent gonna wanna hear this but i truly do not believe this mistake could’ve been made by someone with a phd. after going through brutal peer review several times you just stop doing stuff like this. whoever made this graph clearly has a bachelors degree. maybe a masters
-
GPT-5 hands-on experience and Pareto chart analysis insights
By
–
from being hands on w gpt5 for past 3 weeks + doing this pareto chart for the past 11 months for every lab
-

OpenAI Dominates Intelligence Frontier with Continuously Trained Router Model
By
–
A lot of developers unimpressed with the GPT-5 numbers are missing the bigger picture: @OpenAI now completely dominates the Intelligence Pareto frontier and for the first time, they have officially confirmed a continuously trained realtime router model and provided some
-
SWEbench Scores Inflated by Django Training Data Bias
By
–
good time to remind everyone that a high score on SWEbench really just indicates the training data contained a sufficiently large proportion of Django