Antoine LB
@a.lebaux.co
The tables have turned with claude 4. React's verbose nature is making it harder for humans to follow what the AI is doing.
The question is not "can AI write code" but can humans still follow up
The question is not "can AI write code" but can humans still follow up
July 11, 2025 at 4:15 PM
The tables have turned with claude 4. React's verbose nature is making it harder for humans to follow what the AI is doing.
The question is not "can AI write code" but can humans still follow up
The question is not "can AI write code" but can humans still follow up
Is there a estimated time for when we will be able to use it?
July 10, 2025 at 3:50 PM
Is there a estimated time for when we will be able to use it?
man, I waiting for your video. I'm to dumb to understand a RFC
July 10, 2025 at 1:16 PM
man, I waiting for your video. I'm to dumb to understand a RFC
I don't think so, but if you do I would be happy to try it
July 4, 2025 at 5:29 PM
I don't think so, but if you do I would be happy to try it
Having this type of UI in Obsidian.md would be sick. Maybe a Obsidian plugin
July 4, 2025 at 4:30 PM
Having this type of UI in Obsidian.md would be sick. Maybe a Obsidian plugin
"mom, that is my true self"
June 27, 2025 at 1:52 PM
"mom, that is my true self"
Hope I can see it on his youtube www.youtube.com/@Micode/videos
Micode
Don't have a lot of time to spend on IT? I do.
You might as well enjoy the little things that few people know!
I don't pretend to be training you,
but if I made you curious by sharing something with...
www.youtube.com
June 25, 2025 at 9:51 PM
Hope I can see it on his youtube www.youtube.com/@Micode/videos
Changing the tech stack has a real impact on how much can AI help you.
Now we use Drizzle, so Cursor has a lot more context and is more helpful bc everything is in the project itself.
We used to use StrapiJS for the backend, and it couldn't really get it.
Now we use Drizzle, so Cursor has a lot more context and is more helpful bc everything is in the project itself.
We used to use StrapiJS for the backend, and it couldn't really get it.
June 25, 2025 at 1:06 PM
Changing the tech stack has a real impact on how much can AI help you.
Now we use Drizzle, so Cursor has a lot more context and is more helpful bc everything is in the project itself.
We used to use StrapiJS for the backend, and it couldn't really get it.
Now we use Drizzle, so Cursor has a lot more context and is more helpful bc everything is in the project itself.
We used to use StrapiJS for the backend, and it couldn't really get it.
"The Valyrian test", and we inject the word "dracarys" to see if is able to get it, is around the end of it. And see how bad it hallucinates.
I tested and Google Speech didn't say a word, Gladia did hallucinate
www.youtube.com/watch?v=qd4L...
I tested and Google Speech didn't say a word, Gladia did hallucinate
www.youtube.com/watch?v=qd4L...
Daenerys speaking Valyrian - "Dovaogedys!"
YouTube video by expo
www.youtube.com
June 17, 2025 at 11:20 PM
"The Valyrian test", and we inject the word "dracarys" to see if is able to get it, is around the end of it. And see how bad it hallucinates.
I tested and Google Speech didn't say a word, Gladia did hallucinate
www.youtube.com/watch?v=qd4L...
I tested and Google Speech didn't say a word, Gladia did hallucinate
www.youtube.com/watch?v=qd4L...
Note: is all in spanish :/
Now that I think about it, there are some simple tests:
- silence
- noise
- talk like a human but invent words
It would test the hallucination potential in transcriptions, which is a real issue at least for us.
Now that I think about it, there are some simple tests:
- silence
- noise
- talk like a human but invent words
It would test the hallucination potential in transcriptions, which is a real issue at least for us.
June 17, 2025 at 11:07 PM
Note: is all in spanish :/
Now that I think about it, there are some simple tests:
- silence
- noise
- talk like a human but invent words
It would test the hallucination potential in transcriptions, which is a real issue at least for us.
Now that I think about it, there are some simple tests:
- silence
- noise
- talk like a human but invent words
It would test the hallucination potential in transcriptions, which is a real issue at least for us.
Currently we run Gladia, Google Speech and Open AI in parallel so we have 3 versions of the same audio.
We tried Eleven Labs but had to turn it off bc is hallucinated with silence and it got annoying.
But we constantly switch providers to test around, we want to have the top 3-5 models at all times
We tried Eleven Labs but had to turn it off bc is hallucinated with silence and it got annoying.
But we constantly switch providers to test around, we want to have the top 3-5 models at all times
June 17, 2025 at 10:59 PM
Currently we run Gladia, Google Speech and Open AI in parallel so we have 3 versions of the same audio.
We tried Eleven Labs but had to turn it off bc is hallucinated with silence and it got annoying.
But we constantly switch providers to test around, we want to have the top 3-5 models at all times
We tried Eleven Labs but had to turn it off bc is hallucinated with silence and it got annoying.
But we constantly switch providers to test around, we want to have the top 3-5 models at all times
In case you want to do something more sophisticated as a audio test I work at parlamento.ai where we do live transcriptions of congress. We use multiple transcription services in parallel to get the best result possible.
Happy to share the 1000s of hours of transcriptions in the name of science
Happy to share the 1000s of hours of transcriptions in the name of science
Parlamento AI
Transcripción en tiempo real del Parlamento potenciada por Inteligencia Artificial
parlamento.ai
June 17, 2025 at 10:57 PM
In case you want to do something more sophisticated as a audio test I work at parlamento.ai where we do live transcriptions of congress. We use multiple transcription services in parallel to get the best result possible.
Happy to share the 1000s of hours of transcriptions in the name of science
Happy to share the 1000s of hours of transcriptions in the name of science
I find audio transcription hard to actually test, what would a pelican bike test for audio transcription?
June 17, 2025 at 10:42 PM
I find audio transcription hard to actually test, what would a pelican bike test for audio transcription?
is index.network related to trychroma.com ?
Bc the design of the website is so similar that the background image of index.network has a src="trychroma.com/img/noise.jpg"
Bc the design of the website is so similar that the background image of index.network has a src="trychroma.com/img/noise.jpg"
Chroma
Chroma is the open-source AI application database. Batteries included.
www.trychroma.com
June 17, 2025 at 8:31 PM
is index.network related to trychroma.com ?
Bc the design of the website is so similar that the background image of index.network has a src="trychroma.com/img/noise.jpg"
Bc the design of the website is so similar that the background image of index.network has a src="trychroma.com/img/noise.jpg"
"Telehealth services, including virtual medical care, mental health support, and easy ordering and delivery for prescription medications"
"24/7 roadside assistance through Drive America"
WTF
www.trump.com/media/trump-...
"24/7 roadside assistance through Drive America"
WTF
www.trump.com/media/trump-...
Trump Mobile Launches A Bold New Wireless Service for Americans
The official website of The Trump Organization. Explore our luxury real estate portfolio of the finest hotels, golf courses, estates and more. Learn about our history and the ultimate trump lifestyle.
www.trump.com
June 16, 2025 at 8:42 PM
"Telehealth services, including virtual medical care, mental health support, and easy ordering and delivery for prescription medications"
"24/7 roadside assistance through Drive America"
WTF
www.trump.com/media/trump-...
"24/7 roadside assistance through Drive America"
WTF
www.trump.com/media/trump-...
Thermal images in Memphis. Photograph: Steve Jones/Flight by Southwings for Southern Environmental Law Center.
www.theguardian.com/technology/2...
www.theguardian.com/technology/2...
June 12, 2025 at 5:49 PM
Thermal images in Memphis. Photograph: Steve Jones/Flight by Southwings for Southern Environmental Law Center.
www.theguardian.com/technology/2...
www.theguardian.com/technology/2...
This version looks good enough to be used, the last one didn't
June 11, 2025 at 10:48 PM
This version looks good enough to be used, the last one didn't
Context: After WW2 France also wanted nuclear weapons but could not spend as much as the USA and USSR were spending on development.
The strategy was to develop comercial nuclear applications to justify the development of the military applications.
Not blindly following what the USA and USSR did.
The strategy was to develop comercial nuclear applications to justify the development of the military applications.
Not blindly following what the USA and USSR did.
June 10, 2025 at 6:12 PM
Context: After WW2 France also wanted nuclear weapons but could not spend as much as the USA and USSR were spending on development.
The strategy was to develop comercial nuclear applications to justify the development of the military applications.
Not blindly following what the USA and USSR did.
The strategy was to develop comercial nuclear applications to justify the development of the military applications.
Not blindly following what the USA and USSR did.
"Legal, finance, healthcare, and government professionals get traceable reasoning that meets compliance requirements."
I really like how Mistral tries to create "side products" out of regular LLMs and not just trying to improve reasoning.
It's like when France had to develop nuclear weapons
I really like how Mistral tries to create "side products" out of regular LLMs and not just trying to improve reasoning.
It's like when France had to develop nuclear weapons
June 10, 2025 at 6:06 PM
"Legal, finance, healthcare, and government professionals get traceable reasoning that meets compliance requirements."
I really like how Mistral tries to create "side products" out of regular LLMs and not just trying to improve reasoning.
It's like when France had to develop nuclear weapons
I really like how Mistral tries to create "side products" out of regular LLMs and not just trying to improve reasoning.
It's like when France had to develop nuclear weapons