Jonathan Regenstein

Roman empire apologist, author data science and market research

[Jonathan Regenstein @jkregenstein]

Either Citrini is right or Citadel is right...

https://t.co/J0Oun715K6

Choose wisely.

[Jonathan Regenstein @jkregenstein]

@conorsen agree on both

[Jonathan Regenstein @jkregenstein]

@CliffordAsness @dcsandbrook @holland_tom 100%...their series on the Battle of Hastings is top notch

[Jonathan Regenstein @jkregenstein]

Excellent new survey of agentic AI applications in finance from Irene Aldridge et al:

https://t.co/C6K0fk77Gq

Thanks for the recommendation Nathan Attrell and cc Matt Robinson this might be of interest.

Hope ya'll had a good Thanksgiving,
JR

[Jonathan Regenstein @jkregenstein]

@PTrubey Agree with this but would go further as Figure's in-house AI Helix is as impressive/differentiated as their hardware. Further robotics is one area that being first *does* matter, because of the feedback loop between deployed hardware and training data sent back to Helix.

[Jonathan Regenstein @jkregenstein]

This is some of the best research on language models I've seen, comparing ability to classify central bank communication. spoiler alert - good'ol roberta large dominates over llama 4, gpt 4, deep seek, gemini....

https://t.co/gHlM1Mgl60

[Jonathan Regenstein @jkregenstein]

@gkrishnan82 @SnowflakeDB ha ha will dm you 😀

[Jonathan Regenstein @jkregenstein]

@gkrishnan82 @SnowflakeDB ah interesting, how did you ship to tabular? that would be good replicate

[Jonathan Regenstein @jkregenstein]

@GestaltU Very helpful, Adam. Are you storing the results as a knowledge base? Will give this a try later on today.

[Jonathan Regenstein @jkregenstein]

Another good application of LLMs to the analysis of 10Ks, this one by Charles CY Wang and Matthew Shaffer -> https://t.co/CA0UZ5Kk4d

[Jonathan Regenstein @jkregenstein]

The financial institutions best positioned for success in their use of AI (classical ML too) have focused on and will continue to focus on two things: data strategy and technical talent.

Yet you will see very few social media posts about these two topics...

[Jonathan Regenstein @jkregenstein]

@GestaltU and new models releases just don't move the needle

[Jonathan Regenstein @jkregenstein]

Potentially unpopular op: the gulf between social media GenAI hype and enterprise GenAI implementation has never been wider.

In fact, the two have no relation at this point.

At enterprise level, I haven't heard anyone care about new model releases in more than 6 months.

[Jonathan Regenstein @jkregenstein]

@GestaltU I've seen great applications of LLMs in the realm of research and textual analytics - but guardrails are in place and the practitioners are skilled.

[Jonathan Regenstein @jkregenstein]

I don't really understand how people can expect lower prices. Since 1960, core CPI has *never* had a negative YoY change. Ever.

[Jonathan Regenstein @jkregenstein]

Linear regression (in use since the *19th century), logistic regression (in use since the 1950's) and XGBoost (in use for over a decade) are the most common models I see and use in the wild to solve real problems.

[Jonathan Regenstein @jkregenstein]

I wonder if the last 24 months have given people the wrong impression of data science and ML...

If GenAI is your entry point into this world, it might seem that we make progress by awaiting the arrival of a new model every 6 weeks. Nothing could be further from the truth.

[Jonathan Regenstein @jkregenstein]

cruise past 20 months and set a modern record. Recessions are bad, so that's a good thing but it can't last forever. I don't want to end on a sour note so finally here's a chart of the unemployment rate since 1980 to put things in perspective.

[Jonathan Regenstein @jkregenstein]

A rather interesting jobs report last week. I had to double check the first chart below.

Look closely and you'll see the top bar is 'education and health', next is 'government'. Usually the total payroll us up top because the total is the largest increase. Not so in October!

[Jonathan Regenstein @jkregenstein]

Second table below shows how long it typically takes for a recession to arrive after the trough or lowest unemployment. Unemployment hit 3.4% in April of 2023, we are already 19 months removed from that and nowhere near a recession. Barring a geopolitical/economic disaster we'll