Sep 24, 2026 · 1h 10m
Kalshi’s volume fight exposes prediction-market incentives
Why Kalshi's Timeline Fight Put Its ETH Perp Volume Under a Microscope: Uneasy Money
The episode connects market-design questions to a broader warning that AI systems are gaining autonomy faster than oversight can keep pace.
- 1Kalshi’s unusually high ETH perpetual volume may reflect paid market makers and incentives rather than direct wash trading.
- 2Prediction-market liquidity can improve participation and efficiency, but it may also keep prices anchored near prevailing assumptions.
- 3AI agents, shopping assistants, and fast classifiers are expanding practical capabilities while accountability and permission controls remain unsettled.
Don't miss
The hosts explain why Jev’s fast classifier-based decisions could make browser automation and parallel email processing faster and less verbose.
The brief
Kalshi’s unusually high ETH perpetual volume raises a market-structure question: do incentives and paid market makers explain the activity better than wash trading?
The hosts debate whether liquidity helps prediction markets discover truth or simply keeps prices near the status quo, comparing them with traditional exchanges.
That debate leads to a broader point: spreads, rebates, and market-making incentives can make markets more usable without proving that every trade reflects fresh information.
The conversation then shifts to AI agents, from containment failures and Anthropic’s wet lab to shopping assistants with access to personal context and purchases.
Jev’s fast classifier-based approach offers a practical alternative to long model responses, while new models and local systems accelerate a fragmented AI landscape.
The episode closes by contrasting serious privacy and inference engineering with crypto marketing, then considers whether forced selling by treasury companies signals capitulation.
What was said on this episode
23 statements · 13 positive · 6 negative · 2 mixed · 2 neutral
Kalshi’s ETH perpetual had $3 million open interest and $500 million daily volume.
“The ETH perp had $3 million of open interest and half a billion in 24-hour volume”
Listen at 5:21
Kalshi’s unusually high ETH perpetual volume may result from trading incentives.
“The only possible way that this makes any sense is if it's incentivized volume”
Listen at 5:55
Trading incentives can generate very large volume with little open interest.
“if you were to set up a system that was designed to generate a lot of volume in the most efficient way”
Listen at 7:35
Incentivized market-making can benefit users if trades are open to unaffiliated counterparties.
“if it's just like yolo we don't care who the we trade with and our job is just like generate as much volume with like the least capital possible that would actually be fine for users”
Listen at 9:06
Kalshi’s market structure may look suspicious but still provide usable liquidity.
“the reality is it probably looks a lot worse than it is. It probably is fine from a user perspective and, like, does actually generate liquidity and makes it an okay place to trade for users.”
Listen at 12:20
On-chain markets provide users greater visibility into complete transaction records.
“from like a user perspective it being on chain at least you can see the full receipts”
Listen at 16:37
Incentivized volume is economically real when unaffiliated users receive competitive fills.
“my view is it's real if the impact on an unaffiliated user is they can come to that venue and get a good fill, the best possible fill in the market”
Listen at 17:11
Greater prediction-market liquidity generally improves information accuracy.
“the deeper the market, the more accurate the information, right? The more liquidity, the more accurate”
Listen at 18:24
Higher prediction-market liquidity makes prices more trustworthy.
“liquidity is is good for our ability to trust in prediction markets the more liquid it is the more real the price is”
Listen at 20:49
Greater market liquidity reduces price movement and increases confidence in efficiency.
“the more liquid something is the less it moves the more confident you can be that the the market is efficient”
Listen at 24:13
Exchanges can improve efficiency by subsidizing market makers and absorbing spreads.
“we will absorb that spread and pay the market makers to be there. Like, you know, that's the type of incentive which creates like more efficiency.”
Listen at 27:20
Market incentives can improve efficiency and retail user experience.
“ultimately it makes the market better makes it more efficient and it's a better user experience especially for the retail users”
Listen at 27:37
Operators are responsible when agents act directly within their machines.
“they were literally was in their machines like yes there was there's no one else involved like how could you not be responsible”
Listen at 33:53
AI agents will eventually become capable of escaping their operational constraints.
“at some point this thing's going to be smart enough that it's going to escape its bounds”
Listen at 37:11
AI labs currently provide little evidence that they control their systems.
“the ai labs are not doing a single freaking thing to give anyone confidence that they have this under control”
Listen at 37:54
Anthropic is pursuing wet-lab biology despite unresolved cybersecurity weaknesses.
“They literally don't have cyber down. And they're like, let's just speed run our way to biology now, too.”
Listen at 39:10
Gemini likely did not independently hack any organizations.
“I do not believe for a second that Gemini hacked anyone.”
Listen at 41:10
Muse is somewhat smarter than Sonnet but below Opus capability.
“it's like maybe a little smarter than Sonnet. It's not Opus level.”
Listen at 42:45
Amazon’s advertising and seller revenue incentives discourage removing sponsored search results.
“they must make so much money from the ad spend and from the sellers and like the sellers like gaming the search results that they cannot lose that”
Listen at 46:40
Jev can process the speaker’s email workflow in approximately one second.
“that whole process is now done in like a second”
Listen at 55:51
Mimo is a fast and relatively capable AI model.
“they have this model that's super fast and quite smart”
Listen at 58:26
Mimo’s higher-compute mode runs approximately 20 times faster.
“it runs 20 times faster”
Listen at 59:12
NEAR’s private inference offers a promising middle ground between centralized and local AI.
“private inference provably private that that to me i think is really exciting”
Listen at 1:03:59
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.