John D. Patterson: [email protected]

@Chris_Worsey @the_P_God @karpathy Eh, that won't cut it. Turn it loose on the current market in a live/paper trade account and get back to us. I suspect you'll see a very different picture.

@the_P_God @karpathy Let's see, I'm going to do a 10 year back test next

I took the @karpathy autoresearch loop and pointed it at markets. 25 AI agents debate macro, rates, commodities, sectors, and single stocks daily. Every recommendation scored against real outcomes. Worst agent by rolling Sharpe gets its prompt rewritten by the system. Keep or revert. Same loop, prompts are the weights, Sharpe is the loss function. Trained the agents on 18 months of market data. 378 iterations. 54 prompt modifications, 16 survived. The system learned which agents to trust using Darwinian weights — geopolitical, commodities, and the @BillAckman quality compounder rose to the top. The agents even figured out their own portfolio manager was the weakest link before we did! Deployed the trained agents. +22% in 173 days. Best pick: AVGO at $152, held for +128%. The final prompts are evolutionary products — shaped by market feedback, not human intuition. Now running live with my own capital. github.com/chrisworsey55/… Part hedge fund, part research experiment :)

@ChrisLaubAI "Stop following gurus", "Here's what actually works"

This Stanford paper just proved that 90% of prompt engineering advice is wrong. I spent 6 months testing every "expert" technique. Most of it is folklore. Here's what actually works (backed by real research):

@universeinanegg colab

I am teaching a ~60 person class that involves a lot of Transformers and Language Modeling in the new year. What is the cheapest and easiest solution to getting my students just a bit of compute to play around with?

@karpathy Skeptical view: This is baked into some providers’ tools bc they’re paid by the token. A next area for competition is for one (or more) providers to streamline token usage. As a user, I’m frankly tired of ‘thinking’ models using inane numbers of tokens for a standard response.

I'm noticing that due to (I think?) a lot of benchmarkmaxxing on long horizon tasks, LLMs are becoming a little too agentic by default, a little beyond my average use case. For example in coding, the models now tend to reason for a fairly long time, they have an inclination to start listing and grepping files all across the entire repo, they do repeated web searchers, they over-analyze and over-think little rare edge cases even in code that is knowingly incomplete and under active development, and often come back ~minutes later even for simple queries. This might make sense for long-running tasks but it's less of a good fit for more "in the loop" iterated development that I still do a lot of, or if I'm just looking for a quick spot check before running a script, just in case I got some indexing wrong or made some dumb error. So I find myself quite often stopping the LLMs with variations of "Stop, you're way overthinking this. Look at only this single file. Do not use any tools. Do not over-engineer", etc. Basically as the default starts to slowly creep into the "ultrathink" super agentic mode, I feel a need for the reverse, and more generally good ways to indicate or communicate intent / stakes, from "just have a quick look" all the way to "go off for 30 minutes, come back when absolutely certain".

@DanielJLosey @GaryMarcus PhD level 🤣

my feed right now

The postdoc position in our Cognitive Neuroscience of Creativity Lab at Penn State is now live: tinyurl.com/dkb3bk4z

3 x Funded Ph.D. positions at School of Psychology, University of Plymouth, UK. I am available for PhD supervision in my areas of expertise, as are many of my colleagues. Please RT. jobs.ac.uk/job/DHH203/tea…

Princeton is hiring postdocs interested in projects focused on comparing AI systems to human cognition or using ideas from cognitive science to better understand those systems as part of a new initiative studying Natural and Artificial Minds. Apply at puwebp.princeton.edu/AcadHire/posit…

Honored to have gotten the best poster award at #SfNC2024! Thank you to my advisor @Roger_Beaty for his guidance on this project!

Congratulations to the SfNC2024 Conference Award Winners!!! Best Poster Award#2: Clin Lai 👏👏🥳🥳

Postdoc opportunity! Join the Cognitive Neuroscience of Creativity Lab @PennState for NSF-funded projects on AI, creativity assessment, & neuroimaging. Send CV & research interests to rebeaty@psu.edu. Job ad coming soon. Come do creativity science with us!

@WitteKristin @toby_wise @docqhuys @cpilab Excited to give this a read! I was very surprised by the findings showcased in the tweeprint at first. After further consideration though it made more sense. If the current is bad (anx/depress), what’s the opportunity cost to shifting? Dep on severity, how much worse can it get?

Anxiety and depression might be related to increased, not decreased, exploration behaviour. What could be possible mechanisms underlying this? Beyond thrilled to share my very first first-author preprint with @toby_wise @docqhuys and @cpilab. osf.io/preprints/psya… 🧵⬇️ [1/n]

Congrats to CNCL grad student @PaulVDiStefano3 for his master’s thesis award on creativity and over-inclusive thinking 👏 🎉

Metaphor is abstract and uniquely comprehensible to humans, right? In work led by @PaulVDiStefano3, we found that LLMs could be trained to predict human creativity judgments for metaphors (r >= .7)—on par with arguably less complex/abstract, creativity tasks—check it out!

Just released our latest pre-print and my first first-author paper 'Automatic Scoring of Metaphor Creativity with Large Language Models' Big thanks to my collaborators @Roger_Beaty and @jdpttrsn Check it out here: osf.io/preprints/psya…

You can check out our latest pre-print on automated creativity scoring: psyarxiv.com/g5qvf Big thanks to my co-authors:@RRPcreativity,@Roger_Beaty,@jdpttrsn, @PaulVDiStefano3, Nadine T. Maliakkal

Out now: A huge international team, led by @penn_state's @jdpttrsn and @Roger_Beaty, including @MPI_ae's Julia F. Christensen, has expanded automatic #semantic evaluation methods to further #multilingual #research on #creativity. Now available #openaccess: doi.org/10.1037/aca000…

Dalai Alpaca is here! cocktailpeanut.github.io/dalai Now you can run the Alpaca LLM on your computer (Mac, Windows, Linux) with just ONE command! Best of all, all you need is just around 4.2GB of disk space! Just run this:

Large language models change everything for linguistics, starting with Chomsky. Featuring: LLMs as scientific theories, response to prior takes, "why" questions in language, acquisition... and how the field should have seen this coming. Paper is here: lingbuzz.net/lingbuzz/007180

alpaca-lora: Code for reproducing the Stanford Alpaca InstructLLaMA result on consumer hardware github: github.com/tloen/alpaca-l…