Siddhartha Dalal

Professor, Columbia U., Teach Machine Learning, Cryptocurrency; CTO, Rand Corp; Chief Data Scientist, AIG; Research VP, Xerox; Bellcore/Bell Labs Fellow, Exec D

[Siddhartha Dalal @siddalal]

Check out my latest article: Don’t Let AI Agents Become Your Next Maintenance Sink https://t.co/azC0BzYkLJ via @LinkedIn

[Nemil Dalal @nemild]

Fun family story: my dad (@siddharthadalal) taught me Basic programming at 6, C programming at 12. I went to work in tech.

Then I showed him Bitcoin and Ethereum at 67. He now teaches a class on crypto at Columbia.

[Vishal Misra @vishalmisra]

1/9 🧵 OpenAI's latest model, GPT-4 o1-preview, demonstrates impressive results using Chain of Thought (CoT) reasoning internally. Our recent paper "Beyond the Black Box: A Statistical Model for LLM Reasoning and Inference" explains how and why https://t.co/Zz2fcRRuWx

[Siddhartha Dalal @siddalal]

I just gave a talk at Bangaloru on Artificial General Intelligence with a number of applications in the context of India and Jainism. YouTube link follows. Will appreciate comments. https://t.co/OIgrxKXolU

[Siddhartha Dalal @siddalal]

@agpatriota @vishalmisra @sirbayes There are many ways for constructing priors (e.g., experts-prior experience==training data). So in effect, LLM may be creating prior. By Dirichlet (D) approx any prior on multinomial is a mixture of D, and each D has the behavior of In-Context-Learning. Thus for any prior.

[Siddhartha Dalal @siddalal]

@agpatriota @vishalmisra @sirbayes However, the paper shows that in effect what is generated, as observed in the in-context learning and prompt responses, is consistent with Bayesian Learning with a prior. So it is consistent with implicitly defining prior and updating it.

[Siddhartha Dalal @siddalal]

@agpatriota @vishalmisra @sirbayes As mentioned in the paper, the current generation of LLMs are explicitly using transformers for multinomial probabilities and generating responses. There are other architectures like Mamba that produce similar behavior.

[Vishal Misra @vishalmisra]

New work by us on Large Language Models - how and why they work, and what is “in-context-learning”

We show that (to quote Dave Blei when he saw our work) “ICL is not magic: it is (consistent with) LM smoothing and Bayesian statistics!” (1/n)

https://t.co/Zz2fcRQX6Z

[Rutgers SAS @RutgersSAS]

"I hope the world will get new Gandhis, Mandelas, and Martin Luther Kings," says Siddhartha Dalal, whose gift supports @RutgersU program in Jainism, a faith focused on nonviolence. https://t.co/RZtdtsuWOz @RU_Foundation, @RutgersSASHUM, @RUDiversity, @jainism,

[Siddhartha Dalal @siddalal]

0x21b93e22712f3548543801f49425c958BF8Dcab8

[Siddhartha Dalal @siddalal]

0xA389Ee95698385E2Fb710236BB4C2306145489BD

[Columbia University @Columbia]

Ransomware attacks are becoming increasingly common and are hard to stop because of the pseudo-anonymity of the Bitcoin network. But Professor Siddhartha Dalal is on the case, with new ways to track down bad actors. @ColumbiaSPS https://t.co/W3ahXlgL5a

[Jadson Jall, PhD @JadsonJall]

I ended my day with this excellent conversation about Leadership in a Time of Global Crises: Overcoming Obstacles with Professor Amy Hungerford in conversation with @SidDalal, @BethFishYoshida, @hedhj16, @Revkin, and Arthur Lerner-Lam.
This event was organized by @Columbia_SPS.

[Siddhartha Dalal @siddalal]

In 2016, ALS tragically snatched away my wife Alka's life as it has for many others. Support my first half marathon in NYC to find a cure to ALS: https://t.co/UV190UdCZA

[Siddhartha Dalal @siddalal]

@fedex Failed to deliver a package promised today - forced me to wait whole day; tracking information was wrong. No specific time for tomorrow's delivery. Don't trust delivery date and time. VERY POOR SERVICE!

[Columbia SPS @Columbia_SPS]

Professor of Professional Practice @SidDalal discusses his study measuring media bias with @BusinessInsider:

[Columbia SPS @Columbia_SPS]

Is the media biased against Trump? Researchers @siddalal, Berkay Adlim & @Lesk_M have an answer. Read “How to Measure Relative Bias in Media Coverage” in the October issue of @signmagazine, a joint publication of @AmstatNews and @RoyalStatSoc. https://t.co/xYwTeBBvPy

[Siddhartha Dalal @siddalal]

https://t.co/S3z19fw427
Since the outcomes of quantum computing are probabilistic, it would be great to understand more about uncertainties of the answers.

[Siddhartha Dalal @siddalal]

With the Quantum Advances claimed by Google, what should be done about security for cryptocurrencies? https://t.co/FNkkcVwPhZ

[Siddhartha Dalal @siddalal]

#cryptocurrency I would appreciate material for a lecture on new regulations and laws on cryptocurrencies in my course on Blockchains/Cryptocurrencies/Analytics at Columbia U.