The Transcript
RT @TheTranscript_: $GM CEO on their onshoring plans:
"As we look further ahead, our annual production in the U.S. is expected to rise to an industry-leading 2 million units after we begin production of the Chevrolet Equinox in Kansas, bring the Chevrolet Blazer to Tennessee and add incremental capacity for the Cadillac Escalade and launch our next-generation full-size pickups at Orion Assembly in Michigan.#
tweet
RT @TheTranscript_: $GM CEO on their onshoring plans:
"As we look further ahead, our annual production in the U.S. is expected to rise to an industry-leading 2 million units after we begin production of the Chevrolet Equinox in Kansas, bring the Chevrolet Blazer to Tennessee and add incremental capacity for the Cadillac Escalade and launch our next-generation full-size pickups at Orion Assembly in Michigan.#
tweet
Offshore
Photo
Michael Fritzell (Asian Century Stocks)
RT @smartkarma: S Korea Value Up: more reasons why S Korea will continue to outperform Taiwan and actually make to DM this year. https://t.co/goTEbVeWUL
tweet
RT @smartkarma: S Korea Value Up: more reasons why S Korea will continue to outperform Taiwan and actually make to DM this year. https://t.co/goTEbVeWUL
tweet
Offshore
Photo
Javier Blas
Regular readers know we at @Opinion had flagged BP had far more debt than the company's prefered metric (~$22 bn). Look at net debt + hybrids + leases + off-balance sheet items and it's >$50 bn.
Now, BP acknowledges the issue and promises a hollistic view of debt. About time. https://t.co/MqV5yqOohO
tweet
Regular readers know we at @Opinion had flagged BP had far more debt than the company's prefered metric (~$22 bn). Look at net debt + hybrids + leases + off-balance sheet items and it's >$50 bn.
Now, BP acknowledges the issue and promises a hollistic view of debt. About time. https://t.co/MqV5yqOohO
tweet
Offshore
Photo
Benjamin Hernandez๐
Built this community because trading alone is brutal. My WhatsApp traders get real-time breakout calls, breaking news first, and genuine support from people who actually trade.
Get access โ https://t.co/71FIJId47G
Reply "Hi" immediately for picks.
$HOOD $IREN $OPEN $RKLB $ASTS
tweet
Built this community because trading alone is brutal. My WhatsApp traders get real-time breakout calls, breaking news first, and genuine support from people who actually trade.
Get access โ https://t.co/71FIJId47G
Reply "Hi" immediately for picks.
$HOOD $IREN $OPEN $RKLB $ASTS
โก 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
DAIR.AI
// Automating Sub-Agent Creation for Agentic Orchestration //
Multi-agent systems are powerful but inflexible.
Building agentic systems today relies on static and predefined roles.
For example, an agentic AI coder might have a coder agent, a searcher agent, a reviewer agent.
Each hardwired with fixed capabilities and coordination patterns.
But real-world tasks are dynamic. A predefined set of agents can't cover every subtask that emerges in open-ended environments.
This new research introduces AOrchestra, a framework where a central orchestrator automatically creates specialized sub-agents on the fly.
Every agent is modeled as a simple four-tuple: Instruction, Context, Tools, Model. The orchestrator concretizes this tuple at each step, spawning a tailored executor for each subtask on demand.
How does this work in practice?
The orchestrator doesn't execute tasks itself. It focuses exclusively on decomposing objectives, curating task-relevant context, selecting appropriate tools and models, and delegating execution.
Each sub-agent gets a clean working context with only the information it needs, avoiding the context rot that plagues long-horizon tasks.
The results:
On GAIA, AOrchestra reaches 80.00% pass@1 with Gemini-3-Flash, a 13.94-point absolute jump over OpenHands (66.06%). On Terminal-Bench 2.0, it hits 52.86%, an 18.57-point gain over the strongest baseline. On SWE-Bench-Verified, it achieves 82.00%. Across all three benchmarks, that's a 16.28% average relative improvement.
What makes this especially compelling?
The orchestration itself is learnable. Fine-tuning a Qwen3-8B model as the orchestrator boosts GAIA accuracy by 11.51 points. In-context learning for cost-aware routing improves accuracy by 3.03% while cutting costs by 18.5%, pushing the system toward a Pareto-efficient frontier.
Why it matters?
Instead of manually engineering agent roles, AOrchestra shows that dynamic, on-demand agent creation with a simple compositional abstraction consistently outperforms static multi-agent designs across diverse benchmarks.
Paper: https://t.co/mO7WZOMMI8
Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c
tweet
// Automating Sub-Agent Creation for Agentic Orchestration //
Multi-agent systems are powerful but inflexible.
Building agentic systems today relies on static and predefined roles.
For example, an agentic AI coder might have a coder agent, a searcher agent, a reviewer agent.
Each hardwired with fixed capabilities and coordination patterns.
But real-world tasks are dynamic. A predefined set of agents can't cover every subtask that emerges in open-ended environments.
This new research introduces AOrchestra, a framework where a central orchestrator automatically creates specialized sub-agents on the fly.
Every agent is modeled as a simple four-tuple: Instruction, Context, Tools, Model. The orchestrator concretizes this tuple at each step, spawning a tailored executor for each subtask on demand.
How does this work in practice?
The orchestrator doesn't execute tasks itself. It focuses exclusively on decomposing objectives, curating task-relevant context, selecting appropriate tools and models, and delegating execution.
Each sub-agent gets a clean working context with only the information it needs, avoiding the context rot that plagues long-horizon tasks.
The results:
On GAIA, AOrchestra reaches 80.00% pass@1 with Gemini-3-Flash, a 13.94-point absolute jump over OpenHands (66.06%). On Terminal-Bench 2.0, it hits 52.86%, an 18.57-point gain over the strongest baseline. On SWE-Bench-Verified, it achieves 82.00%. Across all three benchmarks, that's a 16.28% average relative improvement.
What makes this especially compelling?
The orchestration itself is learnable. Fine-tuning a Qwen3-8B model as the orchestrator boosts GAIA accuracy by 11.51 points. In-context learning for cost-aware routing improves accuracy by 3.03% while cutting costs by 18.5%, pushing the system toward a Pareto-efficient frontier.
Why it matters?
Instead of manually engineering agent roles, AOrchestra shows that dynamic, on-demand agent creation with a simple compositional abstraction consistently outperforms static multi-agent designs across diverse benchmarks.
Paper: https://t.co/mO7WZOMMI8
Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c
tweet
Offshore
Video
Dimitry Nakhla | Babylon Capitalยฎ
One reason $V remains one of Chris Hohnโs largest holdings is its network effect โ something thatโs very difficult to replicate at scale.
As Hohn explains:
โPayments is one where weโve been shareholders of Visa a long timeโฆ it has this huge, ever-growing network connecting every customer and every bank globally. As the network grows, it becomes increasingly difficult to replicate.โ
___
๐๐๐ญ๐ฐ๐จ๐ซ๐ค ๐๐๐๐๐๐ญ๐ฌ: ๐ข ๐ฃ๐ถ๐ด๐ช๐ฏ๐ฆ๐ด๐ด ๐ข๐ฅ๐ท๐ข๐ฏ๐ต๐ข๐จ๐ฆ ๐ธ๐ฉ๐ฆ๐ณ๐ฆ ๐ต๐ฉ๐ฆ ๐ท๐ข๐ญ๐ถ๐ฆ ๐ฐ๐ง ๐ต๐ฉ๐ฆ ๐ฑ๐ณ๐ฐ๐ฅ๐ถ๐ค๐ต ๐ฐ๐ณ ๐ด๐ฆ๐ณ๐ท๐ช๐ค๐ฆ ๐ช๐ฏ๐ค๐ณ๐ฆ๐ข๐ด๐ฆ๐ด ๐ข๐ด ๐ฎ๐ฐ๐ณ๐ฆ ๐ถ๐ด๐ฆ๐ณ๐ด ๐ฑ๐ข๐ณ๐ต๐ช๐ค๐ช๐ฑ๐ข๐ต๐ฆ, ๐ฎ๐ข๐ฌ๐ช๐ฏ๐จ ๐ต๐ฉ๐ฆ ๐ฏ๐ฆ๐ต๐ธ๐ฐ๐ณ๐ฌ ๐ฉ๐ข๐ณ๐ฅ๐ฆ๐ณ ๐ต๐ฐ ๐ณ๐ฆ๐ฑ๐ญ๐ข๐ค๐ฆ ๐ฐ๐ณ ๐ณ๐ฆ๐ฑ๐ญ๐ช๐ค๐ข๐ต๐ฆ ๐ฐ๐ท๐ฆ๐ณ ๐ต๐ช๐ฎ๐ฆ.
Visa isnโt just a payments rail โ itโs a global data, trust, and connectivity network built over decades. Every transaction strengthens fraud detection, authorization accuracy, and value-added services, while raising switching costs for banks and merchants.
That scale is the moat. To replicate it, youโd need billions of cards, global merchant acceptance, regulatory trust, and many years of transaction history โ not just better software.
___
It helps explain why $V now represents 15%+ of TCI after Hohn increased his stake ~47% in Q3 2025.
Today both $V (4.10% NTM FCF yield) and $MA (3.71% NTM FCF Yield) trade near the higher end of their historical forward FCF Yield valuation.
Network effects like these compound quietly โ but consistently โ over time.
___
Video: Money Maze Podcast (11/13/2025)
tweet
One reason $V remains one of Chris Hohnโs largest holdings is its network effect โ something thatโs very difficult to replicate at scale.
As Hohn explains:
โPayments is one where weโve been shareholders of Visa a long timeโฆ it has this huge, ever-growing network connecting every customer and every bank globally. As the network grows, it becomes increasingly difficult to replicate.โ
___
๐๐๐ญ๐ฐ๐จ๐ซ๐ค ๐๐๐๐๐๐ญ๐ฌ: ๐ข ๐ฃ๐ถ๐ด๐ช๐ฏ๐ฆ๐ด๐ด ๐ข๐ฅ๐ท๐ข๐ฏ๐ต๐ข๐จ๐ฆ ๐ธ๐ฉ๐ฆ๐ณ๐ฆ ๐ต๐ฉ๐ฆ ๐ท๐ข๐ญ๐ถ๐ฆ ๐ฐ๐ง ๐ต๐ฉ๐ฆ ๐ฑ๐ณ๐ฐ๐ฅ๐ถ๐ค๐ต ๐ฐ๐ณ ๐ด๐ฆ๐ณ๐ท๐ช๐ค๐ฆ ๐ช๐ฏ๐ค๐ณ๐ฆ๐ข๐ด๐ฆ๐ด ๐ข๐ด ๐ฎ๐ฐ๐ณ๐ฆ ๐ถ๐ด๐ฆ๐ณ๐ด ๐ฑ๐ข๐ณ๐ต๐ช๐ค๐ช๐ฑ๐ข๐ต๐ฆ, ๐ฎ๐ข๐ฌ๐ช๐ฏ๐จ ๐ต๐ฉ๐ฆ ๐ฏ๐ฆ๐ต๐ธ๐ฐ๐ณ๐ฌ ๐ฉ๐ข๐ณ๐ฅ๐ฆ๐ณ ๐ต๐ฐ ๐ณ๐ฆ๐ฑ๐ญ๐ข๐ค๐ฆ ๐ฐ๐ณ ๐ณ๐ฆ๐ฑ๐ญ๐ช๐ค๐ข๐ต๐ฆ ๐ฐ๐ท๐ฆ๐ณ ๐ต๐ช๐ฎ๐ฆ.
Visa isnโt just a payments rail โ itโs a global data, trust, and connectivity network built over decades. Every transaction strengthens fraud detection, authorization accuracy, and value-added services, while raising switching costs for banks and merchants.
That scale is the moat. To replicate it, youโd need billions of cards, global merchant acceptance, regulatory trust, and many years of transaction history โ not just better software.
___
It helps explain why $V now represents 15%+ of TCI after Hohn increased his stake ~47% in Q3 2025.
Today both $V (4.10% NTM FCF yield) and $MA (3.71% NTM FCF Yield) trade near the higher end of their historical forward FCF Yield valuation.
Network effects like these compound quietly โ but consistently โ over time.
___
Video: Money Maze Podcast (11/13/2025)
tweet
Offshore
Photo
God of Prompt
creating ads for clients on autopilot
> simple form
> n8n worklow
> nano banana API
added this to my n8n automations bundle! https://t.co/dyPbWqMS5l
tweet
creating ads for clients on autopilot
> simple form
> n8n worklow
> nano banana API
added this to my n8n automations bundle! https://t.co/dyPbWqMS5l
tweet
Illiquid
Option 3. Bought Seikoh Giken more than 2.5x up in 6 months and itโs still gonna double.
tweet
Option 3. Bought Seikoh Giken more than 2.5x up in 6 months and itโs still gonna double.
You buy a stock already up 100% in 6 months and it doubles.
versus
You buy a stock down 50% in 6 months and it doubles.
In which scenario are you more satisfied with yourself? tweet
X (formerly Twitter)
Citrini (@Citrini7) on X
You buy a stock already up 100% in 6 months and it doubles.
versus
You buy a stock down 50% in 6 months and it doubles.
In which scenario are you more satisfied with yourself?
versus
You buy a stock down 50% in 6 months and it doubles.
In which scenario are you more satisfied with yourself?
Offshore
Photo
Dimitry Nakhla | Babylon Capitalยฎ
RT @DividendDynasty: โOver 95% of our revenue is tied to proprietary benchmarks, differentiated data, and critical workflow tools.โ
Just a reminder, S&P Global is not a meme stock folksโฆ
Long $SPGI
tweet
RT @DividendDynasty: โOver 95% of our revenue is tied to proprietary benchmarks, differentiated data, and critical workflow tools.โ
Just a reminder, S&P Global is not a meme stock folksโฆ
Long $SPGI
Buying is an understatement at these levelsโฆ
$SPGI currently -17.5% on the dayโฆ https://t.co/1LAg7z2an6 tweet