Dr. Daniel Bender
@aidfulai.bsky.social
NotebookLM expert ◆ Teaches you knowledge extraction through AI tools ◆ Making information find you ◆ PhD in computer science ◆ Dad
😄🕐 AI Happy Hour: AI news, open-weight AI and more
Looking forward to talking to my AI buddies and everyone else who is interested in these topics. 👇 (link to join in first comment)
🚨 Again on DISCORD instead of an X-space, today at 7 pm (UTC), which is the next full hour!
Looking forward to talking to my AI buddies and everyone else who is interested in these topics. 👇 (link to join in first comment)
🚨 Again on DISCORD instead of an X-space, today at 7 pm (UTC), which is the next full hour!
November 4, 2025 at 7:22 PM
😄🕐 AI Happy Hour: AI news, open-weight AI and more
Looking forward to talking to my AI buddies and everyone else who is interested in these topics. 👇 (link to join in first comment)
🚨 Again on DISCORD instead of an X-space, today at 7 pm (UTC), which is the next full hour!
Looking forward to talking to my AI buddies and everyone else who is interested in these topics. 👇 (link to join in first comment)
🚨 Again on DISCORD instead of an X-space, today at 7 pm (UTC), which is the next full hour!
🚀 Cut latency, ditch the Cloud!
November 1, 2025 at 7:53 PM
🚀 Cut latency, ditch the Cloud!
Greetings from the AIDev 5 (an AI developer meeting) which this time is part of the AI Week Frankfurt event.
Next is a talk from Michael Dietz, Customer Engineer at Google, about AI agents using the MCP and A2A protocols. 🤓
Next is a talk from Michael Dietz, Customer Engineer at Google, about AI agents using the MCP and A2A protocols. 🤓
October 29, 2025 at 12:47 PM
Greetings from the AIDev 5 (an AI developer meeting) which this time is part of the AI Week Frankfurt event.
Next is a talk from Michael Dietz, Customer Engineer at Google, about AI agents using the MCP and A2A protocols. 🤓
Next is a talk from Michael Dietz, Customer Engineer at Google, about AI agents using the MCP and A2A protocols. 🤓
Digitizing printed text via optical character recognition (OCR);
- Classic OCR fails with gibberish and your spellchecker catches it.
- AI OCR fails with plausible words that fit the context perfectly. The error is really hard to catch.
Is the latter really what we want?
- Classic OCR fails with gibberish and your spellchecker catches it.
- AI OCR fails with plausible words that fit the context perfectly. The error is really hard to catch.
Is the latter really what we want?
October 27, 2025 at 10:01 PM
Digitizing printed text via optical character recognition (OCR);
- Classic OCR fails with gibberish and your spellchecker catches it.
- AI OCR fails with plausible words that fit the context perfectly. The error is really hard to catch.
Is the latter really what we want?
- Classic OCR fails with gibberish and your spellchecker catches it.
- AI OCR fails with plausible words that fit the context perfectly. The error is really hard to catch.
Is the latter really what we want?
There is a new star in the world of open models on the horizon:
MiniMax M2
Given that this has less active parameters and is sparser than other open models, the benchmark results are pretty good. 👇
However, you should vibe test it yourself via OpenRouter (currently for free).
MiniMax M2
Given that this has less active parameters and is sparser than other open models, the benchmark results are pretty good. 👇
However, you should vibe test it yourself via OpenRouter (currently for free).
October 25, 2025 at 4:18 PM
There is a new star in the world of open models on the horizon:
MiniMax M2
Given that this has less active parameters and is sparser than other open models, the benchmark results are pretty good. 👇
However, you should vibe test it yourself via OpenRouter (currently for free).
MiniMax M2
Given that this has less active parameters and is sparser than other open models, the benchmark results are pretty good. 👇
However, you should vibe test it yourself via OpenRouter (currently for free).
32 books on AI and humanity for $16. That's the top tier of @humble "The Human Side of AI" collection. It starts at just $1 for 3 books.
The bundle contains @mitpress titles that tackle the question:
How do we understand our relationship with machines? 👇
The bundle contains @mitpress titles that tackle the question:
How do we understand our relationship with machines? 👇
October 24, 2025 at 9:09 PM
32 books on AI and humanity for $16. That's the top tier of @humble "The Human Side of AI" collection. It starts at just $1 for 3 books.
The bundle contains @mitpress titles that tackle the question:
How do we understand our relationship with machines? 👇
The bundle contains @mitpress titles that tackle the question:
How do we understand our relationship with machines? 👇
GLM 4.6 is the most powerful open-weight model for coding.
With 375B params, it's likely too large to run it on your own hardware.
Subscriptions from z. ai start at $2.70 (code in comment) for the first month.
Use it on platforms like Claude Code and OpenCode. 👇
With 375B params, it's likely too large to run it on your own hardware.
Subscriptions from z. ai start at $2.70 (code in comment) for the first month.
Use it on platforms like Claude Code and OpenCode. 👇
October 13, 2025 at 7:37 PM
GLM 4.6 is the most powerful open-weight model for coding.
With 375B params, it's likely too large to run it on your own hardware.
Subscriptions from z. ai start at $2.70 (code in comment) for the first month.
Use it on platforms like Claude Code and OpenCode. 👇
With 375B params, it's likely too large to run it on your own hardware.
Subscriptions from z. ai start at $2.70 (code in comment) for the first month.
Use it on platforms like Claude Code and OpenCode. 👇
🚀 Attention all LLM enthusiasts! 🚀
Leaks share specs of the RTX 5070 Ti Super, the next-gen powerhouse with 24GB VRAM at an unbelievable $800.
Prepare for a game-changing shift in the market for power users handling large language models on their own devices. 👇🧵
Leaks share specs of the RTX 5070 Ti Super, the next-gen powerhouse with 24GB VRAM at an unbelievable $800.
Prepare for a game-changing shift in the market for power users handling large language models on their own devices. 👇🧵
September 4, 2025 at 8:29 AM
🚀 Attention all LLM enthusiasts! 🚀
Leaks share specs of the RTX 5070 Ti Super, the next-gen powerhouse with 24GB VRAM at an unbelievable $800.
Prepare for a game-changing shift in the market for power users handling large language models on their own devices. 👇🧵
Leaks share specs of the RTX 5070 Ti Super, the next-gen powerhouse with 24GB VRAM at an unbelievable $800.
Prepare for a game-changing shift in the market for power users handling large language models on their own devices. 👇🧵
If you don't want Anthropic to use your chats from Claude to train future models, make sure your privacy settings are correct.
Go to settings (1) -> Privacy (2) -> Turn "Help improve Claude" off. 👇
Go to settings (1) -> Privacy (2) -> Turn "Help improve Claude" off. 👇
September 2, 2025 at 3:04 PM
If you don't want Anthropic to use your chats from Claude to train future models, make sure your privacy settings are correct.
Go to settings (1) -> Privacy (2) -> Turn "Help improve Claude" off. 👇
Go to settings (1) -> Privacy (2) -> Turn "Help improve Claude" off. 👇
As everyone explores GPT-5, it reveals a dark truth:
today's AI might disappear tomorrow.
OpenAI's unexpected model replacements show how cloud dependency creates "digital quicksand" for users and businesses.
More about it and a way out in my latest newsletter issue. 👇
today's AI might disappear tomorrow.
OpenAI's unexpected model replacements show how cloud dependency creates "digital quicksand" for users and businesses.
More about it and a way out in my latest newsletter issue. 👇
August 8, 2025 at 10:03 AM
As everyone explores GPT-5, it reveals a dark truth:
today's AI might disappear tomorrow.
OpenAI's unexpected model replacements show how cloud dependency creates "digital quicksand" for users and businesses.
More about it and a way out in my latest newsletter issue. 👇
today's AI might disappear tomorrow.
OpenAI's unexpected model replacements show how cloud dependency creates "digital quicksand" for users and businesses.
More about it and a way out in my latest newsletter issue. 👇
Download the latest Kiro update!
By being an early adopter of Amazon's VS Code fork, Kiro, I kept getting an update alert that annoyingly led me to a waitlist.
But I've discovered a little-known subpage that lets current users directly access Kiro releases for installation or upgrades. 👇
By being an early adopter of Amazon's VS Code fork, Kiro, I kept getting an update alert that annoyingly led me to a waitlist.
But I've discovered a little-known subpage that lets current users directly access Kiro releases for installation or upgrades. 👇
August 5, 2025 at 12:58 PM
Download the latest Kiro update!
By being an early adopter of Amazon's VS Code fork, Kiro, I kept getting an update alert that annoyingly led me to a waitlist.
But I've discovered a little-known subpage that lets current users directly access Kiro releases for installation or upgrades. 👇
By being an early adopter of Amazon's VS Code fork, Kiro, I kept getting an update alert that annoyingly led me to a waitlist.
But I've discovered a little-known subpage that lets current users directly access Kiro releases for installation or upgrades. 👇
Your future self will thank you for learning AI tools when they were still novel.
July 19, 2025 at 2:07 PM
Your future self will thank you for learning AI tools when they were still novel.
Everyone should think about building an AI wrapper for his biggest pain
Skip building your own model. Move fast, grab your niche, and collect the data that'll give you the edge later.
Skip building your own model. Move fast, grab your niche, and collect the data that'll give you the edge later.
July 18, 2025 at 12:08 PM
Everyone should think about building an AI wrapper for his biggest pain
Skip building your own model. Move fast, grab your niche, and collect the data that'll give you the edge later.
Skip building your own model. Move fast, grab your niche, and collect the data that'll give you the edge later.
Kiro, the new Visual Studio Code fork from Amazon has now a waitlist. 🤔
Released on Monday as preview, and so far free to download and register. That changed, likely due to the huge interest of vibecoders who want to test the more structured approach of Kiro.
Released on Monday as preview, and so far free to download and register. That changed, likely due to the huge interest of vibecoders who want to test the more structured approach of Kiro.
July 17, 2025 at 3:36 PM
Kiro, the new Visual Studio Code fork from Amazon has now a waitlist. 🤔
Released on Monday as preview, and so far free to download and register. That changed, likely due to the huge interest of vibecoders who want to test the more structured approach of Kiro.
Released on Monday as preview, and so far free to download and register. That changed, likely due to the huge interest of vibecoders who want to test the more structured approach of Kiro.
● 🚀 Just got Amazon's new coding editor Kiro running on my Arch Linux!
Unlike for Windows, MacOS and Ubuntu, for Arch Kiro is provided as a complete self-contained folder that you simply move to /opt. Everything's included: the editor, dependencies, and even a beautiful icon.
Unlike for Windows, MacOS and Ubuntu, for Arch Kiro is provided as a complete self-contained folder that you simply move to /opt. Everything's included: the editor, dependencies, and even a beautiful icon.
July 16, 2025 at 7:00 PM
● 🚀 Just got Amazon's new coding editor Kiro running on my Arch Linux!
Unlike for Windows, MacOS and Ubuntu, for Arch Kiro is provided as a complete self-contained folder that you simply move to /opt. Everything's included: the editor, dependencies, and even a beautiful icon.
Unlike for Windows, MacOS and Ubuntu, for Arch Kiro is provided as a complete self-contained folder that you simply move to /opt. Everything's included: the editor, dependencies, and even a beautiful icon.
Just discovered Gemini-CLI dropped end of June 👀
Already a fan of Claude Code (now part of the standard Claude subscription), but Gemini Pro’s coding capabilities are also 🔥
Testing both side by side on a new project:
➡️ Extracting my writing style from my Substack newsletter
Already a fan of Claude Code (now part of the standard Claude subscription), but Gemini Pro’s coding capabilities are also 🔥
Testing both side by side on a new project:
➡️ Extracting my writing style from my Substack newsletter
July 10, 2025 at 8:33 AM
Just discovered Gemini-CLI dropped end of June 👀
Already a fan of Claude Code (now part of the standard Claude subscription), but Gemini Pro’s coding capabilities are also 🔥
Testing both side by side on a new project:
➡️ Extracting my writing style from my Substack newsletter
Already a fan of Claude Code (now part of the standard Claude subscription), but Gemini Pro’s coding capabilities are also 🔥
Testing both side by side on a new project:
➡️ Extracting my writing style from my Substack newsletter
Ever wondered how model quantization (FP16, Q8, Q4) *really* affects performance?
There's an analogy that makes the trade-offs crystal clear... and it involves something you might drink. 😉🍺
Kudos to @jtdavies for this brilliant comparison. 🙏
See the image for the full explanation! 👇
There's an analogy that makes the trade-offs crystal clear... and it involves something you might drink. 😉🍺
Kudos to @jtdavies for this brilliant comparison. 🙏
See the image for the full explanation! 👇
June 11, 2025 at 7:12 PM
Ever wondered how model quantization (FP16, Q8, Q4) *really* affects performance?
There's an analogy that makes the trade-offs crystal clear... and it involves something you might drink. 😉🍺
Kudos to @jtdavies for this brilliant comparison. 🙏
See the image for the full explanation! 👇
There's an analogy that makes the trade-offs crystal clear... and it involves something you might drink. 😉🍺
Kudos to @jtdavies for this brilliant comparison. 🙏
See the image for the full explanation! 👇
Sam Altman shares a glimpse of the future he is seeing in front of us. 👇
The changes will be huge, and now is the time to prepare yourself for it by AI-first thinking.
BUT, do it problem centric: Whatever you are working on, think about how AI can support you. 🤔
The changes will be huge, and now is the time to prepare yourself for it by AI-first thinking.
BUT, do it problem centric: Whatever you are working on, think about how AI can support you. 🤔
June 11, 2025 at 7:15 AM
Sam Altman shares a glimpse of the future he is seeing in front of us. 👇
The changes will be huge, and now is the time to prepare yourself for it by AI-first thinking.
BUT, do it problem centric: Whatever you are working on, think about how AI can support you. 🤔
The changes will be huge, and now is the time to prepare yourself for it by AI-first thinking.
BUT, do it problem centric: Whatever you are working on, think about how AI can support you. 🤔
Magistral was released by @mistralai. Their first stinking models. 🎉
You can run locally the small version of Magistral, which builds on top of Mistral Small 3.1 and has 24B parameters.
But why does it not have the tool support in @ollama which it should have? 🥹
You can run locally the small version of Magistral, which builds on top of Mistral Small 3.1 and has 24B parameters.
But why does it not have the tool support in @ollama which it should have? 🥹
June 10, 2025 at 10:05 PM
Magistral was released by @mistralai. Their first stinking models. 🎉
You can run locally the small version of Magistral, which builds on top of Mistral Small 3.1 and has 24B parameters.
But why does it not have the tool support in @ollama which it should have? 🥹
You can run locally the small version of Magistral, which builds on top of Mistral Small 3.1 and has 24B parameters.
But why does it not have the tool support in @ollama which it should have? 🥹
Summrize → Blinkist, but for free
Summrize offers entirely free book summaries of popular nonfiction books. I just checked the summary of the latest book I have read, and I have to say it is quite good.
I'd use it to decide whether to read a book, not instead of the book.
Summrize offers entirely free book summaries of popular nonfiction books. I just checked the summary of the latest book I have read, and I have to say it is quite good.
I'd use it to decide whether to read a book, not instead of the book.
June 6, 2025 at 9:05 PM
Summrize → Blinkist, but for free
Summrize offers entirely free book summaries of popular nonfiction books. I just checked the summary of the latest book I have read, and I have to say it is quite good.
I'd use it to decide whether to read a book, not instead of the book.
Summrize offers entirely free book summaries of popular nonfiction books. I just checked the summary of the latest book I have read, and I have to say it is quite good.
I'd use it to decide whether to read a book, not instead of the book.
🤯 Blown away by FLUX.1 Kontext from @bfl_ml!
I wanted to edit a photo of me for my upcoming newsletter issue (wasn't looking my best! 😉). Used the new Flux model for minor, realistic tweaks. The result? It still looks like me. 👇
Does it look like me to YOU? Let me know! 🤔
I wanted to edit a photo of me for my upcoming newsletter issue (wasn't looking my best! 😉). Used the new Flux model for minor, realistic tweaks. The result? It still looks like me. 👇
Does it look like me to YOU? Let me know! 🤔
June 4, 2025 at 9:49 PM
🤯 Blown away by FLUX.1 Kontext from @bfl_ml!
I wanted to edit a photo of me for my upcoming newsletter issue (wasn't looking my best! 😉). Used the new Flux model for minor, realistic tweaks. The result? It still looks like me. 👇
Does it look like me to YOU? Let me know! 🤔
I wanted to edit a photo of me for my upcoming newsletter issue (wasn't looking my best! 😉). Used the new Flux model for minor, realistic tweaks. The result? It still looks like me. 👇
Does it look like me to YOU? Let me know! 🤔
The likely most convenient way to get the best prices at using generative AI models via API is @openrouterai
On top, they provide nice usage statistics. 👇
- overall usage is going up
- smaller and with that cheaper models are mostly used
- model usage differs along categories
On top, they provide nice usage statistics. 👇
- overall usage is going up
- smaller and with that cheaper models are mostly used
- model usage differs along categories
June 3, 2025 at 2:12 PM
The likely most convenient way to get the best prices at using generative AI models via API is @openrouterai
On top, they provide nice usage statistics. 👇
- overall usage is going up
- smaller and with that cheaper models are mostly used
- model usage differs along categories
On top, they provide nice usage statistics. 👇
- overall usage is going up
- smaller and with that cheaper models are mostly used
- model usage differs along categories
🤔 Noticing some variability in image gen quality in ChatGPT-4o?
Look closely to see if you get an image from the new model or the older DALL·E 3 during peak load. The new model builds the image line by line. 🧐
🚀 EU users: The new image generation is LIVE - no VPN needed! 🙌
Look closely to see if you get an image from the new model or the older DALL·E 3 during peak load. The new model builds the image line by line. 🧐
🚀 EU users: The new image generation is LIVE - no VPN needed! 🙌
April 9, 2025 at 11:42 AM
🤔 Noticing some variability in image gen quality in ChatGPT-4o?
Look closely to see if you get an image from the new model or the older DALL·E 3 during peak load. The new model builds the image line by line. 🧐
🚀 EU users: The new image generation is LIVE - no VPN needed! 🙌
Look closely to see if you get an image from the new model or the older DALL·E 3 during peak load. The new model builds the image line by line. 🧐
🚀 EU users: The new image generation is LIVE - no VPN needed! 🙌
Unlock the Best Open-Source LLM for FREE:
Llama 4 Maverick on OpenRouter - use it via API or directly in their chat interface
Honestly, it's a mystery how it's possible, but you can tap into countless models via OpenRouter's API at no cost. It's absolutely incredible!
Llama 4 Maverick on OpenRouter - use it via API or directly in their chat interface
Honestly, it's a mystery how it's possible, but you can tap into countless models via OpenRouter's API at no cost. It's absolutely incredible!
April 7, 2025 at 7:53 PM
Unlock the Best Open-Source LLM for FREE:
Llama 4 Maverick on OpenRouter - use it via API or directly in their chat interface
Honestly, it's a mystery how it's possible, but you can tap into countless models via OpenRouter's API at no cost. It's absolutely incredible!
Llama 4 Maverick on OpenRouter - use it via API or directly in their chat interface
Honestly, it's a mystery how it's possible, but you can tap into countless models via OpenRouter's API at no cost. It's absolutely incredible!
New issue of my AI newsletter with a focus on privacy-friendly AI which you can run locally:
🤝 AI without surveillance: How local models are changing the game
This issue
- On-device AI is the future
- DeepSeek V3's impressive performance
- Small teams leveraging AI for growth
🤝 AI without surveillance: How local models are changing the game
This issue
- On-device AI is the future
- DeepSeek V3's impressive performance
- Small teams leveraging AI for growth
March 30, 2025 at 9:34 PM
New issue of my AI newsletter with a focus on privacy-friendly AI which you can run locally:
🤝 AI without surveillance: How local models are changing the game
This issue
- On-device AI is the future
- DeepSeek V3's impressive performance
- Small teams leveraging AI for growth
🤝 AI without surveillance: How local models are changing the game
This issue
- On-device AI is the future
- DeepSeek V3's impressive performance
- Small teams leveraging AI for growth