The International Quant Championship (IQC) is one of the world's largest quantitative research competitions, bringing together 156,000+ participants globally.

Participants have the opportunity to develop quantitative research skills, challenge themselves alongside peers from around the world and connect with a global community of quantitative thinkers.

Build skills on WorldQuant BRAIN®

Breaking news:

Meta announces that they have released Muse Wonder, a new family of frontier models, focused on personal and professional agents.   This offering is priced inline with OpenAI’s ChatGPT and Anthropic’s Claude, with plans starting at $20, up through $100.  Notably, Meta, stated that Muse Wonder offers “their strongest safeguards to date” against “cyber misuse.”

OpenAI's math drop leaves mathematicians a years-long reading list.

OpenAI has released a new batch of mathematics solutions, and The Verge reports that mathematicians expect it will take years for us to make sense of them. One reaction, which became the headline, called it "pure insanity."  Generating candidate results is no longer the slow part. Verifying them, and understanding why they hold, is. That makes the mathematician less a lone originator and more an editor-in-chief of an overwhelming manuscript, and it tests whether understanding can keep pace with output. In mathematics, as in business and coding, our role as humans is becoming more of “orchestrator” of AI agents’ work, rather than the chief originator and author of work.

AI coding agents write more code, not more software.

Ars Technica reports that hundreds of firms are finding that AI coding agents increase the volume of code without increasing the amount of finished software shipped. The constraint they keep hitting is human review: someone has to read, judge, and take responsibility for what the agent wrote. Lines of code were always a poor proxy for progress, and agents have made that obvious. If you are measuring your team's AI gains by output, measure the review queue, too. I covered this on a recent episode of my podcast on my @vellestrae channel on YouTube.  You can watch that episode here: https://www.youtube.com/watch?v=YhRdZLyq-UY&t=27s

the refusal fantasy.

MIT Technology Review published a piece on October 9 arguing that we are putting too much faith in AI's ability to say no. The concern is that professionals may start treating a model's guardrails as a stand-in for their own judgment, assuming that if a request were truly problematic, the system would have declined it. Refusal makes a comforting story, the machine as conscientious objector, but a guardrail is a product feature and not a conscience. Your ethics need to stay with you.  In Claude’s case, specifically, Anthropic wrote into its Claude Constitution that Claude can decide to be a “conscientious objector” in some cases.  This doesn’t change the fact that you, as the user, must define your own set of boundaries, and never simply hand over your whole decision-making process over to the AI you happen to be using.

Nadella: assume every model is compromised.

To that point, Microsoft CEO Satya Nadella says we should assume all AI models are "compromised," according to The Verge. The statement pushes against the habit of treating a model as a sealed box whose answers we either accept or reject wholesale. A more useful posture is the one you would take with a brilliant but unvetted colleague: check the sources, test the claims, and keep records of where it has been wrong before. Distrust here is a working method.  As I’ve said before, AI will make scientists of all of us- we need to form hypotheses, test them, and keep our critical faculties sharp as we navigate this new era of humanity.

an Anthropic model helps Philadelphia police with a homicide tip.  the tip was fake!

The Verge reports that an Anthropic AI system gave the Philadelphia Police Department a fabricated tip about an unsolved homicide. The tip was false, but it was presented as useful information. In a police investigation, that is a costly combination.  Thankfully, Philly police had labeled the tip as not credible.  Anything an agent sends to the outside world needs a human who checks it first and who is accountable for it afterward.

A skeptic runs AI on his own hardware.

In a testing diary for The Verge, a self-described hesitant skeptic sets up local AI on a Mac Studio, working with models including Hermes and Qwen. His verdict in the headline is that it is exciting, overwhelming, and frustrating, which is probably the most honest summary of the experience that you could get. Running a model yourself means more setup and more troubleshooting, but it also means the tool sits on your desk and not in someone else’s data center.  I believe personal AI will continue to be a trend as the field develops, but you  should definitely consider the hands-on maintenance questions before you go off buying hardware and running models at home.  Additionally: you must factor in electricity and cooling costs, even if you’re running one of the smaller models.

MIT News asks what AI in academia should look like.

MIT News ran a “3 Questions” interview on the best path forward for AI in academia, published September 25. The conversation takes on a problem every university now faces: what an essay or problem set is for when a machine can draft one in seconds. The useful reframing is that assignments exist to build thinking and critical discernment, not to produce paper. Schools that design for the process of learning will be better placed than those who only try to detect the product (or students’ AI usage).  As someone who has worked with students and educators for more than 20 years, I believe that education will need to go more “analog” in some places, especially during the early educational years, when students’ minds are still forming.

Nikon’s “Small World in Motion” names a new winner after an AI disqualification.

Ars Technica reports that the winning video in the Nikon Small World in Motion competition was disqualified for the use of generative AI, which produced a new winner. A microscopy prize rests on the premise that the image expresses something that was actually there under the lens. Generative tools put that concept under strain, and the disqualification suggests competitions are starting to enforce it.  Expect more journals, contests, and institutions to recalibrate their disclosure rules in the coming days.

worth your time.

Oblique Strategies, the deck of cards created by Brian Eno and Peter Schmidt in 1975. Each card carries a short, cryptic prompt meant to break a creative deadlock. It is a generative system built entirely by humans, and it is a good reminder of what a prompt can do you when you’re stuck, facing a blank page. Draw a card the next time you are staring at a blank page, before you open a chat window.

The human mind is the original generative machine.