John Stewart
@johnstewart.dev
Web Platform at MLB ⚾️
AI Enthusiast 🤖
Writing about AI, web dev, and more on my newsletter 🥞The Breakfast Dev☕️
Newsletter: https://www.breakfast.dev
Website: https://www.johnstewart.dev
AI Enthusiast 🤖
Writing about AI, web dev, and more on my newsletter 🥞The Breakfast Dev☕️
Newsletter: https://www.breakfast.dev
Website: https://www.johnstewart.dev
Asked Claude 3.7 to build my son a game that it created.
Almost nailed it on the first try. All the balls were in the top left corner.
After some nudging it got it.
Exciting to see that my son and I can create near infinite games for ourselves. Games special to us.
Almost nailed it on the first try. All the balls were in the top left corner.
After some nudging it got it.
Exciting to see that my son and I can create near infinite games for ourselves. Games special to us.
February 27, 2025 at 2:13 PM
Asked Claude 3.7 to build my son a game that it created.
Almost nailed it on the first try. All the balls were in the top left corner.
After some nudging it got it.
Exciting to see that my son and I can create near infinite games for ourselves. Games special to us.
Almost nailed it on the first try. All the balls were in the top left corner.
After some nudging it got it.
Exciting to see that my son and I can create near infinite games for ourselves. Games special to us.
Export controls are still the best tool we have when it comes to preventing China from winning the AI race.
January 29, 2025 at 10:51 PM
Export controls are still the best tool we have when it comes to preventing China from winning the AI race.
R1 isn't as interesting as V3. It's using RL like o1
January 29, 2025 at 10:51 PM
R1 isn't as interesting as V3. It's using RL like o1
DeepSeek's cost for the ~50k Hopper chips is ~$1B, which isn't far off from US companies.
January 29, 2025 at 10:51 PM
DeepSeek's cost for the ~50k Hopper chips is ~$1B, which isn't far off from US companies.
All of this is expected progress on the same curve. What's different is the fact that China produced this.
January 29, 2025 at 10:51 PM
All of this is expected progress on the same curve. What's different is the fact that China produced this.
DeepSeek did not train as good a model for $6M. Also we got some insight into the true cost of Claude 3.5 Sonnet at "a few $10Ms"
January 29, 2025 at 10:51 PM
DeepSeek did not train as good a model for $6M. Also we got some insight into the true cost of Claude 3.5 Sonnet at "a few $10Ms"
DeepSeek V3 was the real innovation (a month ago). Specifically they innovated on an area called the KV cache using a method called "mixture of experts".
January 29, 2025 at 10:51 PM
DeepSeek V3 was the real innovation (a month ago). Specifically they innovated on an area called the KV cache using a method called "mixture of experts".
When it comes to shifting the paradigm he discusses that in 2024 the real shift has been adding in reinforcement learning (RL) on top of pre-trained models.
January 29, 2025 at 10:51 PM
When it comes to shifting the paradigm he discusses that in 2024 the real shift has been adding in reinforcement learning (RL) on top of pre-trained models.
The 3 dynamics of AI development
- Scaling laws
- Shifting the curve
- Shifting the paradigm
- Scaling laws
- Shifting the curve
- Shifting the paradigm
January 29, 2025 at 10:51 PM
The 3 dynamics of AI development
- Scaling laws
- Shifting the curve
- Shifting the paradigm
- Scaling laws
- Shifting the curve
- Shifting the paradigm
China is here and they are messing things up for the US in a big way. DeepSeek R-1 was the first reasoning model they dropped. Then yesterday they released their Jasper models. They open sourced everything. US AI companies suggested China was 5 years behind - this is clearly not true anymore.
January 28, 2025 at 3:54 PM
China is here and they are messing things up for the US in a big way. DeepSeek R-1 was the first reasoning model they dropped. Then yesterday they released their Jasper models. They open sourced everything. US AI companies suggested China was 5 years behind - this is clearly not true anymore.
Reasoning Models are here and OpenAI isn't the only game in town anymore. DeepSeek R-1 is sending shock waves through the world right now not just for their research but US AI companies are now being devalued because of this. AGI will come from these models, so expect a lot more of these in 2025.
January 28, 2025 at 3:54 PM
Reasoning Models are here and OpenAI isn't the only game in town anymore. DeepSeek R-1 is sending shock waves through the world right now not just for their research but US AI companies are now being devalued because of this. AGI will come from these models, so expect a lot more of these in 2025.
Reasoning Models are here and OpenAI isn't the only game in town anymore. DeepSeek R-1 is sending shock waves through the world right now not just for their research but US AI companies are now being devalued because of this. AGI will come from these models, so expect a lot more of these in 2025.
January 28, 2025 at 3:45 PM
Reasoning Models are here and OpenAI isn't the only game in town anymore. DeepSeek R-1 is sending shock waves through the world right now not just for their research but US AI companies are now being devalued because of this. AGI will come from these models, so expect a lot more of these in 2025.
There is actually nothing intelligent about this feature
January 13, 2025 at 2:42 PM
There is actually nothing intelligent about this feature
Example conversation between malicious user request and model
January 10, 2025 at 2:00 PM
Example conversation between malicious user request and model
Apple Playground result side by side
January 10, 2025 at 12:52 AM
Apple Playground result side by side
I should go back to church
December 22, 2024 at 1:52 AM
I should go back to church
Where is the Oregon Trail floppy?!
Miss these days
Miss these days
December 1, 2024 at 10:38 PM
Where is the Oregon Trail floppy?!
Miss these days
Miss these days