The Alternative Data Podcast

There are genuine nuggets - particularly the phased evolution of alt-data alpha and the Telegram sentiment research finding - but they're surrounded by lengthy biographical small-talk, visa discussions, and South Africa anecdotes that add no operational value. The insight-to-filler ratio is poor for a 48-minute runtime.

“in the early days, I think especially that example I mentioned where I found a data set, you simply use the insight and you can immediately harvest it. For alpha, that went away relatively quickly”

“the sentiment itself was not predictive. Positive sentiment lagged the price by a few hours and so it's really of no use. Um, negative sentiment was coincident with the price.”

The Telegram crypto sentiment findings (sentiment lagged, message volume more predictive) are a genuinely specific research observation, and the framing of AI as enabling scrape maintenance at broader scale is fresh. But most of the episode recycles well-worn takes about AI, social media data, and junior talent concerns.

“Positive sentiment lagged the price by a few hours and so it's really of no use. Um, negative sentiment was coincident with the price. So the moment the price drops, negative sentiment spikes.”

“if agents are able to touch it then you're probably also able to analyze it more systematically. So that's sort of a dual one”

Nico Smuts is a genuine practitioner who went from fundamental equity long-short to senior data scientist at Citadel, with real independent research on Telegram crypto data along the way. However, the non-compete structurally prevents him from discussing the most valuable five years of his career, severely limiting the depth he can offer.

“I was approached by a recruiter on behalf of Citadel and made um, an offer to join the team there”

“I mined that data set, um, extracted about a million messages over uh, nine months and did some analysis on that”

There are some concrete data points - a million Telegram messages over nine months, Europe-to-US-east-coast container costs doubling, ICE launching a futures market, three-year history threshold - but the Citadel constraint forces vagueness on the most interesting material, and many claims about AI and social media are left at the assertion level.

“the cost of sending a container from Europe to US east coast has doubled”

“I extracted about a million messages over uh, nine months and did some analysis on that”

The host occasionally shows craft - asking whether Nico backed the Telegram research with his own money is a sharp follow-up, and he probes the three-year history claim - but large swaths of the episode are consumed by biographical chat, visa logistics, and in-jokes that eat time without generating insight. Vague answers about AI and social media go largely unchallenged.

“I think it's a great innovation. I'm beginning to see why you, why you wound up at Citadel”

“Are you worried that there are huge things uh, moves being made in AI and you're not currently seeing them happen”