Michael Fritzell (Asian Century Stocks)
"Buy SaaS. Not overpriced. Mostly moats."
tweet
"Buy SaaS. Not overpriced. Mostly moats."
https://t.co/KRlTbtnA8A tweet
X (formerly Twitter)
Finbarr Taylor (@finbarr) on X
In Defense of SaaS
Offshore
Photo
Michael Fritzell (Asian Century Stocks)
The bull case for Midea
tweet
The bull case for Midea
South Korea uses more industrial robots per worker than any other country—
This chart shows one way to compare automated manufacturing across countries — it plots the number of robots per 1,000 manufacturing employees.
The chart shows very large differences between countries. South Korea stands out, with more than one robot for every ten manufacturing workers.
Singapore comes second, and China ranks third, close to Germany. The United States sits in the middle, close to the European average, below Switzerland, Denmark and Slovenia.
This perspective shows industrial robot adoption in relative terms. In another Data Insight, I looked at robot adoption in absolute terms. From that perspective, China stands out by a large margin: it’s a large economy with a huge manufacturing sector, and it has by far the largest stock of industrial robots.
Much of this expansion has happened recently: China’s annual installations increased 12-fold over a decade, helping it catch up to South Korea in terms of robots per worker.
(This Data Insight was written by @EOrtizOspina.) tweet
Offshore
Photo
Michael Fritzell (Asian Century Stocks)
White collars workers will use AI tools to become better at their jobs. And then the competitive frontier will shift to design, writing, sales, access to data, etc. Some will increase their output and the number of customers they serve. Others will serve customers differently.
tweet
White collars workers will use AI tools to become better at their jobs. And then the competitive frontier will shift to design, writing, sales, access to data, etc. Some will increase their output and the number of customers they serve. Others will serve customers differently.
Are you sposed to clap for this? https://t.co/h6SPD6KWsI tweet
Moon Dev
i used six claude codes all day today
i was using six claude codes all day and my efficiency is through the roof
if you want to see how to actually step on the gas watch the replay
grab a ticket for tomorrows zoom and get the api key there too
join here: https://t.co/Aw7dcEvv2n
moondev
tweet
i used six claude codes all day today
i was using six claude codes all day and my efficiency is through the roof
if you want to see how to actually step on the gas watch the replay
grab a ticket for tomorrows zoom and get the api key there too
join here: https://t.co/Aw7dcEvv2n
moondev
tweet
Offshore
Photo
God of Prompt
RT @godofprompt: Claude is insane for product management.
I reverse-engineered how top PMs at Google, Meta, and Anthropic use it.
The difference is night and day.
Here are 10 prompts they don't want you to know (but I'm sharing anyway): https://t.co/7RApvBHQ66
tweet
RT @godofprompt: Claude is insane for product management.
I reverse-engineered how top PMs at Google, Meta, and Anthropic use it.
The difference is night and day.
Here are 10 prompts they don't want you to know (but I'm sharing anyway): https://t.co/7RApvBHQ66
tweet
Michael Fritzell (Asian Century Stocks)
RT @blondesnmoney: One of the sharpest small cap guys, @GuastyWinds , is finally back on Twitter. Made a bunch from his excellent callouts in 2023 and 2024, really excited to see what he has next and highly recommend a follow.
tweet
RT @blondesnmoney: One of the sharpest small cap guys, @GuastyWinds , is finally back on Twitter. Made a bunch from his excellent callouts in 2023 and 2024, really excited to see what he has next and highly recommend a follow.
tweet
Offshore
Photo
Jukan
I don’t have a girlfriend, but I still wanted to try some chocolate, so I looked around online.
And… it turns out Louis Vuitton makes chocolate too? It’s offline-exclusive, though.
The price is a whopping $360. I have no idea what kind of ridiculous money flex that is. https://t.co/I1YviRZ9Do
tweet
I don’t have a girlfriend, but I still wanted to try some chocolate, so I looked around online.
And… it turns out Louis Vuitton makes chocolate too? It’s offline-exclusive, though.
The price is a whopping $360. I have no idea what kind of ridiculous money flex that is. https://t.co/I1YviRZ9Do
tweet
Michael Fritzell (Asian Century Stocks)
RT @david_katunaric: @MoS_Investing To a man with an AI everything looks like disruption
tweet
RT @david_katunaric: @MoS_Investing To a man with an AI everything looks like disruption
tweet
anon
Yeah, but Ichiyoshi January 16, 2025 report: "AIMECHATEC: Initiating strong buy on anticipated long-term growth in semiconductor bonders and debonders".
Ishibashi-san did a long report on both AIMECHATEC and Shikoku Kasei (CCL sector report) before becoming baggers.
tweet
Yeah, but Ichiyoshi January 16, 2025 report: "AIMECHATEC: Initiating strong buy on anticipated long-term growth in semiconductor bonders and debonders".
Ishibashi-san did a long report on both AIMECHATEC and Shikoku Kasei (CCL sector report) before becoming baggers.
This tweet had 95 impressions and four likes. Ofc, the stock is up 124% since then. tweet
X (formerly Twitter)
Illiquid (@illyquid) on X
This tweet had 95 impressions and four likes. Ofc, the stock is up 124% since then.
Offshore
Photo
DAIR.AI
RT @dair_ai: // Improving Efficiency of Evolutionary AI Agents //
Evolutionary AI agents are powerful but can be wasteful.
Systems, inspired by AlphaEvolve and OpenEvolve, iteratively generate, mutate, and refine candidate solutions using LLMs. However, every refinement step invokes the same large model regardless of task difficulty.
Most mutations don't need a 32B model.
This new research introduces AdaptEvolve, a framework that dynamically selects which model handles each evolutionary step based on intrinsic generation confidence.
Instead of routing everything through the largest available model, a lightweight decision tree router estimates whether the small model's output is sufficient or needs escalation.
The confidence signal comes from four entropy-based metrics computed on the small model's token probabilities: Mean Confidence for global assurance, Lowest Group Confidence for localized reasoning collapses, Tail Confidence for solution stability, and Bottom-K% Confidence for distinguishing noise from systematic hallucination.
A shallow decision tree, bootstrapped from just 50 warm-up examples, uses these signals to make real-time routing decisions.
What makes this practical?
The router adapts online. An Adaptive Hoeffding Tree continuously updates its decision boundaries as the evolutionary population drifts toward harder edge cases.
On LiveCodeBench, AdaptEvolve retains 97.9% of the 32B upper-bound accuracy (73.6% vs 75.2%) while cutting compute cost by 34.4%. On MBPP, the router identifies that 85% of queries are solvable by the 4B model alone, reducing cost by 41.5% while maintaining 97.1% of peak accuracy. Across benchmarks, the method reduces total inference compute by 37.9% while retaining 97.5% of the upper-bound performance.
Evolutionary agents don't need maximum capability at every step. Confidence-driven routing turns the cost-capability trade-off from a fixed choice into a dynamic, per-step decision.
Paper: https://t.co/YSNCKZuTeN
Learn to build effective AI Agents in our academy: https://t.co/LRnpZN7L4c
tweet
RT @dair_ai: // Improving Efficiency of Evolutionary AI Agents //
Evolutionary AI agents are powerful but can be wasteful.
Systems, inspired by AlphaEvolve and OpenEvolve, iteratively generate, mutate, and refine candidate solutions using LLMs. However, every refinement step invokes the same large model regardless of task difficulty.
Most mutations don't need a 32B model.
This new research introduces AdaptEvolve, a framework that dynamically selects which model handles each evolutionary step based on intrinsic generation confidence.
Instead of routing everything through the largest available model, a lightweight decision tree router estimates whether the small model's output is sufficient or needs escalation.
The confidence signal comes from four entropy-based metrics computed on the small model's token probabilities: Mean Confidence for global assurance, Lowest Group Confidence for localized reasoning collapses, Tail Confidence for solution stability, and Bottom-K% Confidence for distinguishing noise from systematic hallucination.
A shallow decision tree, bootstrapped from just 50 warm-up examples, uses these signals to make real-time routing decisions.
What makes this practical?
The router adapts online. An Adaptive Hoeffding Tree continuously updates its decision boundaries as the evolutionary population drifts toward harder edge cases.
On LiveCodeBench, AdaptEvolve retains 97.9% of the 32B upper-bound accuracy (73.6% vs 75.2%) while cutting compute cost by 34.4%. On MBPP, the router identifies that 85% of queries are solvable by the 4B model alone, reducing cost by 41.5% while maintaining 97.1% of peak accuracy. Across benchmarks, the method reduces total inference compute by 37.9% while retaining 97.5% of the upper-bound performance.
Evolutionary agents don't need maximum capability at every step. Confidence-driven routing turns the cost-capability trade-off from a fixed choice into a dynamic, per-step decision.
Paper: https://t.co/YSNCKZuTeN
Learn to build effective AI Agents in our academy: https://t.co/LRnpZN7L4c
tweet