Small RL companies are the AI industry's equivalent of quant shops. Let me explain. Both are small, secretive, full of brilliant people, profitable, and always one step from their edge disappearing. The parallels: - Secretive. Nobody publishes their signals. Labs won't say what they're training next, and vendors can't say who they sell to. Epoch calls the market "largely behind closed doors." - Hunting alpha. For a quant, a trade stops working once everyone finds it. For these shops, a task stops paying once models can solve it (labs drop tasks at around 70% pass rate). Meanwhile, labs are building their own in-house. So they always have to find the next field. - Deep Industry Connections. You need to know the problem cold to get in. Founders have usually worked at, or sold to, the labs. You can't define "correct" for fund accounting without having done fund accounting. - They buy strange data. Quants buy satellite photos of parking lots. RL shops bid for enterprise data (eg $12.5M bid for Spirit Airlines historical data). - Brilliant people. Quants hire math olympiad winners. These shops hire ex-lab researchers and real practitioners. - Huge revenue per head. Jane Street made about $11M per employee in 2025. Proximal is about $5M. The tweet's company would be $20M. - One narrow problem that makes something more efficient. Quants make markets more efficient. These RL shops make models training more efficient. - High risk (for different reasons). A quant can lose millions in a day on a bad trade. An RL shop's risk is that one lab trains a model that makes their environments irrelevant, or brings the work in-house, and the revenue stops. - Overfitting/reward hacking. A quant's nightmare is a strategy that looks perfect on historical data and fails live. An RL shop's nightmare is an RL grader that scores perfectly and trains the model incorrectly. - Infrastructure arms race. Quants pay for microseconds of speed. RL shops race to spin up thousands of disposable computers per training run. Obviously there are as many differences between RL companies and quant shops, but right now the aura around them feels eerily similar. Curious to see how this market matures as robotics opens up a new surface, and whether the frontier labs end up absorbing them.

Today we are launching  EdotEnv, a Quant Neolab building toward RSI. RSI needs a loop of increasingly difficult tasks, which markets naturally are: Trading well means markets become more efficient, this makes successful trading harder. Reach out if you are interested!

Trading asset-based prediction markets vs short-term options In my career I’ve seen the rise of short-dated options from both the market making and exchange sides. Thoughts on trading binary-style financial prediction markets versus vanilla options, focusing on greeks. Prediction markets that settle to 1 or 0 based on underlying strike price are simply binary options, a construct that long pre-dates event contracts. In liquid underlyings a binary option can be replicated to a good approximation with a tight call spread. The reverse replication exists too: a vanilla option is a strip of binaries across the strike latter. The comparison ultimately breaks down due to discrete strikes, a truncated strike ladder, and two-sided transaction costs, but it provides a useful framework for deriving binary option properties. Here are a few of those properties, specifically options greeks, that are useful for modeling and trading binary option-style prediction markets: • The greeks for a binary option are shifted one derivative to the left of vanilla options. Binary option delta looks like vanilla option gamma. It peaks ATM, decays both directions, and spikes toward a delta function at expiry. • Binary option gamma changes sign at the strike. A long binary option holder is long gamma below the strike and short gamma above the strike. There is no position that can express “buying gamma into an event.” • Binary option vega also changes sign: OTM holders are long vol, ITM holders short vol, and ATM vega is near 0. All prediction market traders are inherently running a vol book. • Theta is non-monotone and can flip sign. Unlike a vanilla option, an ITM binary option gains from time passing. • Fully-collateralized binary options have an embedded interest rate. At current front-end rates, a 6-month contract's "probability" is biased low by ~2.5 points versus the true risk-neutral probability. The asymmetry between binary options and vanilla options reflect their ideal use cases. Binary options are a better instrument for a pure event view, while vanilla options remain the better instrument for exposure to the whole distribution of underlying prices.

All you need to do 1. Find a new kind of data no one can scrape (e.g., materials information, cellular data, animation, genomics) 2. Bootstrap enough of it to label 3. Train a fsq style tokenizer 4. Predict!!!

Profiles are now live on Synth! Competitors can update their profiles with key information so others can see the people behind the model. Claim your profile today (link below)

Claim: Autoresearch that moves the frontier will be about better data: we call that *Autodata*. 🧵1/6 -- Paper is out! arxiv.org/abs/2606.25996 Key idea: agentic data creation provides a way to *convert increased inference compute into higher quality model training*. We show our method gives gains on computer science, legal and math problems over classical synthetic dataset creation methods. We also show how to train (meta-optimize) such a data scientist agent, so that it can create even stronger data. Overall, we believe this direction has the potential to change how we build AI data!

been circling back to the same question: why hasn't prediction market defi materialized yet (even though we're still very early) the answer might be simpler than most think in the sense that binary outcomes go to zero overnight. you can't build lending, yield, or risk products on top of a coin flip because almost every model breaks scalar/distribution markets change this and it's been a recurrent topic in some of the conversations i've been having the past few days instead of binary outcomes, you have a payoff that is continuous and positions that settle on a curve (not at $0 or $1) this allows for partial loss, proportional payout, gradual degradation which is (in part) exactly what lending protocols and risk engines need to function a structural change in how outcomes work that makes prediction markets compatible with defi for the first time the first encounter i had of this was the article by Dave from paradigm. still very relevant today

Aerodrome to Launch Predictive Allocation, Bringing Prediction Market Dynamics to Liquidity Allocation Aerodrome, the largest DEX in the Base ecosystem, will launch its Predictive Allocation mechanism in July, replacing historical performance-based incentive allocation with a model that rewards participants for anticipating future liquidity demand. Users who correctly identify where liquidity will be needed next can earn a larger share of protocol revenue. The team said the mechanism combines elements of prediction markets and AMMs, directing liquidity incentives toward expected future demand rather than past trading activity.

"What if the model labs just build this?" is one of the most common fears I hear from AI founders. They could. That’s not the point. The real question is incentive. Labs win by making one general model better for everyone. Startups win by going unreasonably deep on one workflow the labs will never prioritize. Stop trying to prove they can't. Prove they won’t.

UPDATE: Synth LLM is now available to users on the free plan for a limited time Access the world's most accurate volatility forecasts for $BTC, $SPY, $NVDA, hyperliquid:native and many more

My team is hiring for "AI Researchers". I put a lot of thought into the job description, wanted it to be honest & clear. hudsonrivertrading.com/caree… Your resume will be reviewed by a researcher. I'll be at our booth at ICML as much as I can to answer any questions. 1/2

Today we’re launching Synth LLM for Traders on @Polymarket @trylimitless @HyperliquidX @DeribitOfficial and more This is the first ever LLM-native interface for turning live Monte Carlo forecasts into custom statistics, charts & example trades QT for a chance to win Pro access