J. Mario Meissner
@jmariomeissner.bsky.social
AI/ML engineer obsessed with AI productivity and coding tools.
Cursor fixed one of my two major complaints I had about them. But... it's a big mess. 🧵
Cursor version 0.50 fixed MAX model pricing by finally billing it against your monthly request quota. And it's billed per million tokens, so you pay exactly what you use.
Cursor version 0.50 fixed MAX model pricing by finally billing it against your monthly request quota. And it's billed per million tokens, so you pay exactly what you use.
May 12, 2025 at 2:40 AM
Cursor fixed one of my two major complaints I had about them. But... it's a big mess. 🧵
Cursor version 0.50 fixed MAX model pricing by finally billing it against your monthly request quota. And it's billed per million tokens, so you pay exactly what you use.
Cursor version 0.50 fixed MAX model pricing by finally billing it against your monthly request quota. And it's billed per million tokens, so you pay exactly what you use.
I'm surprised NotebookLM hasn't been axed by Google yet... Instead, it's getting a dedicated app? 🤯
May 8, 2025 at 12:03 AM
I'm surprised NotebookLM hasn't been axed by Google yet... Instead, it's getting a dedicated app? 🤯
Gemini 2.5 Pro is my go-to life advisor. I frequently give it my whole notes folder and ask it to help me organize my life, give me feedback, plan my next steps, etc.
OpenAI o3 would just tell me what a mess I am. I lacks the human touch, the social skills.
OpenAI o3 would just tell me what a mess I am. I lacks the human touch, the social skills.
May 7, 2025 at 3:04 PM
Gemini 2.5 Pro is my go-to life advisor. I frequently give it my whole notes folder and ask it to help me organize my life, give me feedback, plan my next steps, etc.
OpenAI o3 would just tell me what a mess I am. I lacks the human touch, the social skills.
OpenAI o3 would just tell me what a mess I am. I lacks the human touch, the social skills.
🤯 Blows my mind how they just casually dropped an update to Gemini that again leaps past everyone else in capabilities. Not even a new model release, just a minor update to the existing model.
May 7, 2025 at 1:06 AM
🤯 Blows my mind how they just casually dropped an update to Gemini that again leaps past everyone else in capabilities. Not even a new model release, just a minor update to the existing model.
I want to love Gemini 2.5 Pro in Cursor more. But currently it just doesn't call tools enough, and randomly stops without finishing the task too often.
I'm sure these issues will be fixed over time though.
I'm sure these issues will be fixed over time though.
April 27, 2025 at 1:56 PM
I want to love Gemini 2.5 Pro in Cursor more. But currently it just doesn't call tools enough, and randomly stops without finishing the task too often.
I'm sure these issues will be fixed over time though.
I'm sure these issues will be fixed over time though.
The strongest AI coding workflow right now:
1️⃣ Ask o3 for an extensive implementation plan, covering structure, details and code snippets for difficult parts, leveraging web search and reasoning.
2️⃣ Provide it as context in Cursor w/ Sonnet 3.7 for the actual code implementation.
3️⃣ 💰 Profit.
1️⃣ Ask o3 for an extensive implementation plan, covering structure, details and code snippets for difficult parts, leveraging web search and reasoning.
2️⃣ Provide it as context in Cursor w/ Sonnet 3.7 for the actual code implementation.
3️⃣ 💰 Profit.
April 26, 2025 at 12:04 AM
The strongest AI coding workflow right now:
1️⃣ Ask o3 for an extensive implementation plan, covering structure, details and code snippets for difficult parts, leveraging web search and reasoning.
2️⃣ Provide it as context in Cursor w/ Sonnet 3.7 for the actual code implementation.
3️⃣ 💰 Profit.
1️⃣ Ask o3 for an extensive implementation plan, covering structure, details and code snippets for difficult parts, leveraging web search and reasoning.
2️⃣ Provide it as context in Cursor w/ Sonnet 3.7 for the actual code implementation.
3️⃣ 💰 Profit.
Something weird’s happening with OpenAI’s new o-series… In some benchmarks, o3-mini-high underperforms o3-mini 🤯 It's like the model gaslights itself or spirals into flawed reasoning.
More thinking is not always the right choice?
More thinking is not always the right choice?
April 25, 2025 at 3:00 PM
Something weird’s happening with OpenAI’s new o-series… In some benchmarks, o3-mini-high underperforms o3-mini 🤯 It's like the model gaslights itself or spirals into flawed reasoning.
More thinking is not always the right choice?
More thinking is not always the right choice?
Ultra-quirky little Claude Sonnet secret! 🙈
To control the level of thinking of Sonnet 3.7, you can simply use keywords such as "think" or "think harder". But did you know how to trigger the maximum amount of thinking? Apparently, the magic word is "ultrathink"!
To control the level of thinking of Sonnet 3.7, you can simply use keywords such as "think" or "think harder". But did you know how to trigger the maximum amount of thinking? Apparently, the magic word is "ultrathink"!
April 24, 2025 at 12:03 AM
Ultra-quirky little Claude Sonnet secret! 🙈
To control the level of thinking of Sonnet 3.7, you can simply use keywords such as "think" or "think harder". But did you know how to trigger the maximum amount of thinking? Apparently, the magic word is "ultrathink"!
To control the level of thinking of Sonnet 3.7, you can simply use keywords such as "think" or "think harder". But did you know how to trigger the maximum amount of thinking? Apparently, the magic word is "ultrathink"!
Felt like yesterday when there was only one reasonable choice of model in Cursor: Sonnet 3.5.
Now we got so many great choices... what is everyone going with these days?
Now we got so many great choices... what is everyone going with these days?
April 23, 2025 at 3:01 PM
Felt like yesterday when there was only one reasonable choice of model in Cursor: Sonnet 3.5.
Now we got so many great choices... what is everyone going with these days?
Now we got so many great choices... what is everyone going with these days?
Someone explain to me why they called it o4-mini and not o3-mini.
April 20, 2025 at 7:58 AM
Someone explain to me why they called it o4-mini and not o3-mini.
OpenAI literally openly admits that their o3/o4-mini models hallucinate more than o1 🤯
In their publicly available o3/o4-mini model card report, section 3.3, they write that o4-mini hallucinated almost 50% of the time in a specific benchmark, much higher than o1.
In their publicly available o3/o4-mini model card report, section 3.3, they write that o4-mini hallucinated almost 50% of the time in a specific benchmark, much higher than o1.
April 20, 2025 at 12:00 AM
OpenAI literally openly admits that their o3/o4-mini models hallucinate more than o1 🤯
In their publicly available o3/o4-mini model card report, section 3.3, they write that o4-mini hallucinated almost 50% of the time in a specific benchmark, much higher than o1.
In their publicly available o3/o4-mini model card report, section 3.3, they write that o4-mini hallucinated almost 50% of the time in a specific benchmark, much higher than o1.
AI won't take jobs.
Engineers using AI will.
Engineers using AI will.
January 15, 2025 at 4:00 PM
AI won't take jobs.
Engineers using AI will.
Engineers using AI will.
If you visit the Cursor forums, it's been full of people complaining about the quality of LLM answers becoming worse and worse.
There's some truth to it because Cursor tried to be more conservative with context.
But overall, Cursor keeps delivering top notch code for me.
There's some truth to it because Cursor tried to be more conservative with context.
But overall, Cursor keeps delivering top notch code for me.
January 15, 2025 at 10:00 AM
If you visit the Cursor forums, it's been full of people complaining about the quality of LLM answers becoming worse and worse.
There's some truth to it because Cursor tried to be more conservative with context.
But overall, Cursor keeps delivering top notch code for me.
There's some truth to it because Cursor tried to be more conservative with context.
But overall, Cursor keeps delivering top notch code for me.
I use @mantinedev for all of my projects
@tailwindcss and @shadcn are great tools, but after using Mantine I can't go back...
Here's a list of the most common components that I immediately miss when trying to switch 🧵
@tailwindcss and @shadcn are great tools, but after using Mantine I can't go back...
Here's a list of the most common components that I immediately miss when trying to switch 🧵
January 14, 2025 at 2:00 AM
I use @mantinedev for all of my projects
@tailwindcss and @shadcn are great tools, but after using Mantine I can't go back...
Here's a list of the most common components that I immediately miss when trying to switch 🧵
@tailwindcss and @shadcn are great tools, but after using Mantine I can't go back...
Here's a list of the most common components that I immediately miss when trying to switch 🧵
Really enjoyed this podcast episode of Disrupting Japan with @jad_tarifi, going into the weeds of the philosophical aspects of AGI and how transformers might fall short in the near future.
January 13, 2025 at 10:00 AM
Really enjoyed this podcast episode of Disrupting Japan with @jad_tarifi, going into the weeds of the philosophical aspects of AGI and how transformers might fall short in the near future.
✔️ get some fresh air and touch some grass
January 13, 2025 at 8:00 AM
✔️ get some fresh air and touch some grass
Don’t use o1 like a chat model
January 12, 2025 at 10:00 AM
Don’t use o1 like a chat model
I've had a new side project idea for a while, and this weekend I finally started building.
I'm going to work in building-marketing cycles.
2 weeks of building, 2 weeks of outreach and sales.
This marks the start of the first building cycle. Heads down writing code!
I'm going to work in building-marketing cycles.
2 weeks of building, 2 weeks of outreach and sales.
This marks the start of the first building cycle. Heads down writing code!
January 12, 2025 at 2:00 AM
I've had a new side project idea for a while, and this weekend I finally started building.
I'm going to work in building-marketing cycles.
2 weeks of building, 2 weeks of outreach and sales.
This marks the start of the first building cycle. Heads down writing code!
I'm going to work in building-marketing cycles.
2 weeks of building, 2 weeks of outreach and sales.
This marks the start of the first building cycle. Heads down writing code!
General purpose AI agent builders won't break into the larger market.
People want highly purposeful tools specifically made for the problem they have, so they'd rather use a niche AI tool rather than a generalist agent builder.
Huge opportunity for indie hackers.
People want highly purposeful tools specifically made for the problem they have, so they'd rather use a niche AI tool rather than a generalist agent builder.
Huge opportunity for indie hackers.
January 11, 2025 at 4:00 PM
General purpose AI agent builders won't break into the larger market.
People want highly purposeful tools specifically made for the problem they have, so they'd rather use a niche AI tool rather than a generalist agent builder.
Huge opportunity for indie hackers.
People want highly purposeful tools specifically made for the problem they have, so they'd rather use a niche AI tool rather than a generalist agent builder.
Huge opportunity for indie hackers.
I've looked up to indie hackers and solopreneurs for many years now.
Having the courage and skills to not only build, but also approach customers and sell a product by themselves.
This year my goal is to finally earn a dollar as an indie hacker.
Having the courage and skills to not only build, but also approach customers and sell a product by themselves.
This year my goal is to finally earn a dollar as an indie hacker.
January 11, 2025 at 10:00 AM
I've looked up to indie hackers and solopreneurs for many years now.
Having the courage and skills to not only build, but also approach customers and sell a product by themselves.
This year my goal is to finally earn a dollar as an indie hacker.
Having the courage and skills to not only build, but also approach customers and sell a product by themselves.
This year my goal is to finally earn a dollar as an indie hacker.
My prediction on AI for video games:
We're going to continue shifting more computation from rendering to upscaling and generation. The role of game engines is going to shift towards only performing game logic, physics, and general artistic control, leaving everything else to the AI model.
We're going to continue shifting more computation from rendering to upscaling and generation. The role of game engines is going to shift towards only performing game logic, physics, and general artistic control, leaving everything else to the AI model.
January 11, 2025 at 2:00 AM
My prediction on AI for video games:
We're going to continue shifting more computation from rendering to upscaling and generation. The role of game engines is going to shift towards only performing game logic, physics, and general artistic control, leaving everything else to the AI model.
We're going to continue shifting more computation from rendering to upscaling and generation. The role of game engines is going to shift towards only performing game logic, physics, and general artistic control, leaving everything else to the AI model.
Anthropic has a page with tons of great prompts for different use-cases 📝
January 10, 2025 at 4:00 PM
Anthropic has a page with tons of great prompts for different use-cases 📝
I am strongly rooting for Tokyo to become an international center for AI and startups. It's definitely getting better really fast. We even have Waymo coming here soon.
January 10, 2025 at 10:00 AM
I am strongly rooting for Tokyo to become an international center for AI and startups. It's definitely getting better really fast. We even have Waymo coming here soon.
I buy mostly world index funds in my investment accounts.
The S&P 500 has been performing above the world average for a while, and sometimes I feel FOMO for those who go all in on that.
But I have conviction that the S&P 500 boom won't last forever, and I'll be better positioned once that happens.
The S&P 500 has been performing above the world average for a while, and sometimes I feel FOMO for those who go all in on that.
But I have conviction that the S&P 500 boom won't last forever, and I'll be better positioned once that happens.
January 10, 2025 at 2:00 AM
I buy mostly world index funds in my investment accounts.
The S&P 500 has been performing above the world average for a while, and sometimes I feel FOMO for those who go all in on that.
But I have conviction that the S&P 500 boom won't last forever, and I'll be better positioned once that happens.
The S&P 500 has been performing above the world average for a while, and sometimes I feel FOMO for those who go all in on that.
But I have conviction that the S&P 500 boom won't last forever, and I'll be better positioned once that happens.
Haystack, a good LangGraph alternative for building AI agents.
January 9, 2025 at 4:00 PM
Haystack, a good LangGraph alternative for building AI agents.