Offshore
Photo
DAIR.AI
What if you could get multi-agent performance from a single model?
Multi-agent debate systems are powerful. Multiple LLMs can critique each other's reasoning, catch errors, and converge on better answers.
However, the cost scales linearly with the number of agents. Five agents means 5x the compute. Twenty agents means 20x and so on.
But the intelligence gained from debate doesn't have to stay locked behind a compute wall.
This new research introduces AgentArk, a framework that distills the reasoning capabilities of multi-agent debate into a single LLM through trajectory extraction and targeted fine-tuning.
This work addresses an important problem: multi-agent systems are effective but expensive at inference time. AgentArk moves that cost to training time, letting a single model carry the reasoning depth of an entire agent team.
The key idea: run multi-agent debate offline to generate high-quality reasoning traces, then train a smaller model to internalize those patterns.
Five agents debate, one student learns.
AgentArk tests three distillation methods. RSFT uses supervised fine-tuning on correct trajectories. DA filters for diverse reasoning paths. PAD, their strongest method, preserves the full structure of multi-agent deliberation, capturing how agents verify intermediate steps and localize errors.
The results across 120 experiments:
> PAD achieves a 4.8% average gain over single-agent baselines, with in-domain improvements reaching up to 30%. On reasoning quality metrics,
> PAD scores highest in intermediate verification (4.07 vs 2.41 baseline) and reasoning coherence (3.96 vs 1.88 baseline).
>The distilled models also transfer: trained on math, they improve on TruthfulQA with ROUGE-L jumping from 0.613 to 0.657.
Scaling from Qwen3-32B teachers down to Qwen3-0.6B students, the framework holds up. Even sub-billion parameter models absorb meaningful reasoning improvements from multi-agent debate.
Paper: https://t.co/cyPTig221s
Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c
tweet
What if you could get multi-agent performance from a single model?
Multi-agent debate systems are powerful. Multiple LLMs can critique each other's reasoning, catch errors, and converge on better answers.
However, the cost scales linearly with the number of agents. Five agents means 5x the compute. Twenty agents means 20x and so on.
But the intelligence gained from debate doesn't have to stay locked behind a compute wall.
This new research introduces AgentArk, a framework that distills the reasoning capabilities of multi-agent debate into a single LLM through trajectory extraction and targeted fine-tuning.
This work addresses an important problem: multi-agent systems are effective but expensive at inference time. AgentArk moves that cost to training time, letting a single model carry the reasoning depth of an entire agent team.
The key idea: run multi-agent debate offline to generate high-quality reasoning traces, then train a smaller model to internalize those patterns.
Five agents debate, one student learns.
AgentArk tests three distillation methods. RSFT uses supervised fine-tuning on correct trajectories. DA filters for diverse reasoning paths. PAD, their strongest method, preserves the full structure of multi-agent deliberation, capturing how agents verify intermediate steps and localize errors.
The results across 120 experiments:
> PAD achieves a 4.8% average gain over single-agent baselines, with in-domain improvements reaching up to 30%. On reasoning quality metrics,
> PAD scores highest in intermediate verification (4.07 vs 2.41 baseline) and reasoning coherence (3.96 vs 1.88 baseline).
>The distilled models also transfer: trained on math, they improve on TruthfulQA with ROUGE-L jumping from 0.613 to 0.657.
Scaling from Qwen3-32B teachers down to Qwen3-0.6B students, the framework holds up. Even sub-billion parameter models absorb meaningful reasoning improvements from multi-agent debate.
Paper: https://t.co/cyPTig221s
Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c
tweet
Offshore
Photo
Quiver Quantitative
JUST IN: Someone on Polymarket has bet $100K that the US will strike Iran today.
They will win $4,000,000 if it happens.
Insider or gambler? https://t.co/p70QMgWPo1
tweet
JUST IN: Someone on Polymarket has bet $100K that the US will strike Iran today.
They will win $4,000,000 if it happens.
Insider or gambler? https://t.co/p70QMgWPo1
tweet
Offshore
Photo
God of Prompt
RT @rryssf_: MIT researchers just mass-published evidence that the next paradigm after reasoning models isn't bigger context windows ☠️
Recursive Language Models (RLMs) let the model write code to examine, decompose, and recursively call itself over its own input.
the results are genuinely wild. here's the full breakdown:
tweet
RT @rryssf_: MIT researchers just mass-published evidence that the next paradigm after reasoning models isn't bigger context windows ☠️
Recursive Language Models (RLMs) let the model write code to examine, decompose, and recursively call itself over its own input.
the results are genuinely wild. here's the full breakdown:
tweet
Offshore
Video
Bourbon Capital
$AMZN AWS CEO Matt Garman: Utilities don't scale with the level that we need them to scale....so there's probably some steps between here and there where you're going to have to do behind the meter power......we're going to have to fund that ramp up while the world catches up
Congrats Utilities.... you've unlocked a long term Godfather
tweet
$AMZN AWS CEO Matt Garman: Utilities don't scale with the level that we need them to scale....so there's probably some steps between here and there where you're going to have to do behind the meter power......we're going to have to fund that ramp up while the world catches up
Congrats Utilities.... you've unlocked a long term Godfather
tweet
Offshore
Photo
Dimitry Nakhla | Babylon Capital®
Kensho is one of the more under-the-radar assets inside $SPGI.
It’s $SPGI AI and data analytics platform, built to analyze massive, complex datasets across economics, geopolitics, financial markets, and corporate fundamentals—used by asset managers, banks, governments, and enterprises.
While most of the focus is on ratings and indices, Kensho quietly expands $SPGI moat by embedding AI-driven insights deeper into client workflows, increasing switching costs and long-term relevance.
Not flashy, but the kind of capability that can strengthen a toll-booth business over time.
___
3Q 2025 Earnings Call:
“We acquired Kensho back in 2018, including that acquisition since 2018, we have invested over $1B in AI innovation across three developmental stages…
Importantly, our AI innovation serves as a powerful example of our ability to leverage our scale, our expertise and our fiscal discipline. The fact that we made such bold investments early on means that we’ve been able to innovate very efficiently from a financial perspective.”
___
@contextinvestor thank you for the thoughtful comment.
tweet
Kensho is one of the more under-the-radar assets inside $SPGI.
It’s $SPGI AI and data analytics platform, built to analyze massive, complex datasets across economics, geopolitics, financial markets, and corporate fundamentals—used by asset managers, banks, governments, and enterprises.
While most of the focus is on ratings and indices, Kensho quietly expands $SPGI moat by embedding AI-driven insights deeper into client workflows, increasing switching costs and long-term relevance.
Not flashy, but the kind of capability that can strengthen a toll-booth business over time.
___
3Q 2025 Earnings Call:
“We acquired Kensho back in 2018, including that acquisition since 2018, we have invested over $1B in AI innovation across three developmental stages…
Importantly, our AI innovation serves as a powerful example of our ability to leverage our scale, our expertise and our fiscal discipline. The fact that we made such bold investments early on means that we’ve been able to innovate very efficiently from a financial perspective.”
___
@contextinvestor thank you for the thoughtful comment.
I don’t think many investors truly appreciate how deep the moats at $SPGI and $MCO really are.
These aren’t just data businesses — they’re embedded gatekeepers in global capital markets, with network effects, regulatory reliance, & decades of trust that are hard to replicate. tweet
Offshore
Photo
Benjamin Hernandez😎
Stop over-analyzing every ticker. 📊
Analysis paralysis is a profit killer. I provide clean, simple breakout alerts on WhatsApp so beginners and pros alike can act with total confidence.
Get in✅ https://t.co/71FIJId47G
Keep your trading simple and effective.
$SOFI $HOOD $PLTR
tweet
Stop over-analyzing every ticker. 📊
Analysis paralysis is a profit killer. I provide clean, simple breakout alerts on WhatsApp so beginners and pros alike can act with total confidence.
Get in✅ https://t.co/71FIJId47G
Keep your trading simple and effective.
$SOFI $HOOD $PLTR
⚡ The "Electronic Giant" Choice
Recommendation: $AXTI ~$28.20
AXT Inc. is a "Buy" rated powerhouse with a $1.56B valuation. Today's +17.19% rally is backed by a massive 6.97M shares traded.
Reason calling it: High institutional turnover at $28.20 suggests a long-term bottom. https://t.co/dGsp8x98EG tweet
Offshore
Photo
God of Prompt
Prompting is foundational infrastructure, not a surface-level trick - it only appears to die because success means integration. https://t.co/KidFQLFwnO
tweet
Prompting is foundational infrastructure, not a surface-level trick - it only appears to die because success means integration. https://t.co/KidFQLFwnO
tweet
Offshore
Photo
Javier Blas
RT @JavierBlas: COLUMN: Britrish oil giant BP should suspend its $750 million quarterly share buybacks to give incoming CEO Meg O'Neill (who arrives in April) extra financial breathing room.
@Opinion $BP
https://t.co/23vNRD4Nsw
tweet
RT @JavierBlas: COLUMN: Britrish oil giant BP should suspend its $750 million quarterly share buybacks to give incoming CEO Meg O'Neill (who arrives in April) extra financial breathing room.
@Opinion $BP
https://t.co/23vNRD4Nsw
tweet
Offshore
Video
Moon Dev
The Code Equalizer: Build a Professional Market Maker Bot and Stop Trading Manually
imagine a world where the market crashing actually puts more money in your pocket than a moonshot ever could. this isn't some pipe dream but the reality of a market maker that trades algorithmically every single second of the year. if you've ever felt the sting of a liquidation then you know exactly why manual trading is a rigged game that we are finally going to beat
the secret to directionless profit isn't about guessing the next candle but about becoming the liquidity that everyone else is desperate to buy. within the next few minutes you'll see why most bots fail and how a simple kill switch is the only thing standing between you and a total account wipeout. it's time to stop gambling and start building systems that don't care about the news or the hype
i spent hundreds of thousands on developers for different apps in the past because i was convinced i could never code myself. i believed that programming was for geniuses and that i was destined to lose money through overtrading and painful liquidations. then i realized that code is the great equalizer because it removes the emotion that usually destroys a trader's bankroll
now we are here with fully automated systems trading for me while i live my life. i decided to learn to code live on youtube to show that anyone can iterate their way to success if they just stay consistent. the path from getting liquidated to building a market maker is paved with many small failures and constant adjustments
market making is the adult version of a video game where the goal is to win regardless of which way the player moves. if the price goes up we win and if the price goes down we win and if the price goes sideways we also win. we are looking for steady times to buy and sell because that is where the real money is hidden from the average retail trader
to get started we need to connect to an exchange like phemex using the ccxt library in python. this connection is the bridge between our logic and the actual capital in the market. we set our symbol to ubtc and establish our initial inputs like position size and sleep timers to keep things moving smoothly
most people get excited and jump straight into the trading logic but that is how you lose everything. before we ever place a trade we have to implement a kill switch that monitors our risk levels in real time. if our position size ever exceeds a certain limit like one thousand dollars the bot will immediately close everything out
this safety feature is what i call a size kill and it protects you from those rare moments when the bot goes crazy or the market moves too fast. i would much rather have ten thousand small trades at five hundred dollars than one massive trade that ruins the account. risk management is the boring part that actually makes you wealthy over the long term
once the safety nets are in place we dive into the bid and ask data to see where the market is actually priced. we pull the highest bid and the lowest ask to determine where our orders should sit on the books. our bot is designed to be patient and wait for the market to come to us instead of chasing price like a desperate amateur
but how do we know if the market is too volatile to trade in the first place. this is where we bring in indicators like the average true range to measure the volatility of the assets we are watching. if the volatility is too high then our bot will simply sit on the sidelines and wait for things to calm down
we set up a no trade rule where if the current atr is higher than our predefined threshold we won't enter the market. this keeps us out of dangerous flash crashes where the spread becomes unpredictable and the risk of loss increases. we are looking for a steady environment where we can safely provide liquidity and collect our profits
every twenty seconds our bot loops through the logic to check our open positions and active orders. it pulls the latest ca[...]
The Code Equalizer: Build a Professional Market Maker Bot and Stop Trading Manually
imagine a world where the market crashing actually puts more money in your pocket than a moonshot ever could. this isn't some pipe dream but the reality of a market maker that trades algorithmically every single second of the year. if you've ever felt the sting of a liquidation then you know exactly why manual trading is a rigged game that we are finally going to beat
the secret to directionless profit isn't about guessing the next candle but about becoming the liquidity that everyone else is desperate to buy. within the next few minutes you'll see why most bots fail and how a simple kill switch is the only thing standing between you and a total account wipeout. it's time to stop gambling and start building systems that don't care about the news or the hype
i spent hundreds of thousands on developers for different apps in the past because i was convinced i could never code myself. i believed that programming was for geniuses and that i was destined to lose money through overtrading and painful liquidations. then i realized that code is the great equalizer because it removes the emotion that usually destroys a trader's bankroll
now we are here with fully automated systems trading for me while i live my life. i decided to learn to code live on youtube to show that anyone can iterate their way to success if they just stay consistent. the path from getting liquidated to building a market maker is paved with many small failures and constant adjustments
market making is the adult version of a video game where the goal is to win regardless of which way the player moves. if the price goes up we win and if the price goes down we win and if the price goes sideways we also win. we are looking for steady times to buy and sell because that is where the real money is hidden from the average retail trader
to get started we need to connect to an exchange like phemex using the ccxt library in python. this connection is the bridge between our logic and the actual capital in the market. we set our symbol to ubtc and establish our initial inputs like position size and sleep timers to keep things moving smoothly
most people get excited and jump straight into the trading logic but that is how you lose everything. before we ever place a trade we have to implement a kill switch that monitors our risk levels in real time. if our position size ever exceeds a certain limit like one thousand dollars the bot will immediately close everything out
this safety feature is what i call a size kill and it protects you from those rare moments when the bot goes crazy or the market moves too fast. i would much rather have ten thousand small trades at five hundred dollars than one massive trade that ruins the account. risk management is the boring part that actually makes you wealthy over the long term
once the safety nets are in place we dive into the bid and ask data to see where the market is actually priced. we pull the highest bid and the lowest ask to determine where our orders should sit on the books. our bot is designed to be patient and wait for the market to come to us instead of chasing price like a desperate amateur
but how do we know if the market is too volatile to trade in the first place. this is where we bring in indicators like the average true range to measure the volatility of the assets we are watching. if the volatility is too high then our bot will simply sit on the sidelines and wait for things to calm down
we set up a no trade rule where if the current atr is higher than our predefined threshold we won't enter the market. this keeps us out of dangerous flash crashes where the spread becomes unpredictable and the risk of loss increases. we are looking for a steady environment where we can safely provide liquidity and collect our profits
every twenty seconds our bot loops through the logic to check our open positions and active orders. it pulls the latest ca[...]