Eschnou
@eschnou.com
Software Engineer | Post-AGI philosopher | Solo builder
#indieweb #indieai #web3
#indieweb #indieai #web3
AGI or not, meaning of not, if it can (or if management decides it can) displace a large part of the workforce, it seems to me important enough to warrant a societal debate on what we do about it.
June 5, 2025 at 6:17 PM
AGI or not, meaning of not, if it can (or if management decides it can) displace a large part of the workforce, it seems to me important enough to warrant a societal debate on what we do about it.
What if it isn't? What if the impact is real (I believe it, I'm impacted).
Shouldn't we discuss how society as a whole benefits from AI and not just a handful billionaire getting richer on the back of everyone else?
We are frogs in boiling water and will realize too late what we let happen.
Shouldn't we discuss how society as a whole benefits from AI and not just a handful billionaire getting richer on the back of everyone else?
We are frogs in boiling water and will realize too late what we let happen.
June 5, 2025 at 5:22 PM
What if it isn't? What if the impact is real (I believe it, I'm impacted).
Shouldn't we discuss how society as a whole benefits from AI and not just a handful billionaire getting richer on the back of everyone else?
We are frogs in boiling water and will realize too late what we let happen.
Shouldn't we discuss how society as a whole benefits from AI and not just a handful billionaire getting richer on the back of everyone else?
We are frogs in boiling water and will realize too late what we let happen.
How are you so sure? We are already seeing the effects in the creative, consulting, and software industries. At my level, I'm 5~8x productive at writing code. I don't see the new jobs coming at the same rate. Seems sensible to me to approach this with caution and ask the hard questions.
May 30, 2025 at 4:45 PM
How are you so sure? We are already seeing the effects in the creative, consulting, and software industries. At my level, I'm 5~8x productive at writing code. I don't see the new jobs coming at the same rate. Seems sensible to me to approach this with caution and ask the hard questions.
The one to watch is not the cat, but the 30Y treasury. Spiking again. Getting dangerously close to 5%. The King played the (Sorcerer) Apprentice but is losing control of the spell. What if the chain reaction can't be stopped anymore? Once broken, trust is so hard to rebuild. 🎬📈🍿
April 10, 2025 at 4:25 PM
The one to watch is not the cat, but the 30Y treasury. Spiking again. Getting dangerously close to 5%. The King played the (Sorcerer) Apprentice but is losing control of the spell. What if the chain reaction can't be stopped anymore? Once broken, trust is so hard to rebuild. 🎬📈🍿
I'm building/experimenting with such an architecture in this project: github.com/eschnou/Open...
The most amazing bit is seeing the agent react to an error (e.g. a failed web search) and finding other means to achieve its goal.
The most amazing bit is seeing the agent react to an error (e.g. a failed web search) and finding other means to achieve its goal.
April 9, 2025 at 10:32 AM
I'm building/experimenting with such an architecture in this project: github.com/eschnou/Open...
The most amazing bit is seeing the agent react to an error (e.g. a failed web search) and finding other means to achieve its goal.
The most amazing bit is seeing the agent react to an error (e.g. a failed web search) and finding other means to achieve its goal.
On the open source side, I had really good results with qwen-2.5-30b. I think however that you need more than a model. You still need a proper Agentic architecture, where things like chain-of-thought, ReAct, memory, tool use, etc.. remains external to the model.
April 9, 2025 at 10:32 AM
On the open source side, I had really good results with qwen-2.5-30b. I think however that you need more than a model. You still need a proper Agentic architecture, where things like chain-of-thought, ReAct, memory, tool use, etc.. remains external to the model.
Which is not a surprise since it was the stated goal from the start: depreciating the USD to make it more competitive. The challenge is to do this while keeping its status as a reserve currency and not blowing up everything.
At least that's how I understand what is going on.
At least that's how I understand what is going on.
April 9, 2025 at 10:26 AM
Which is not a surprise since it was the stated goal from the start: depreciating the USD to make it more competitive. The challenge is to do this while keeping its status as a reserve currency and not blowing up everything.
At least that's how I understand what is going on.
At least that's how I understand what is going on.
One thing to understand is that all these countries with trade surplus (in $) were recycling that surplus by buying bonds and stocks from the US. With tariffs, less trade, less surplus, less buyers, thus risks of popping the USD bubble.
April 9, 2025 at 10:26 AM
One thing to understand is that all these countries with trade surplus (in $) were recycling that surplus by buying bonds and stocks from the US. With tariffs, less trade, less surplus, less buyers, thus risks of popping the USD bubble.
Thanks @timkellogg.me 🙏 - MCP truly shines in these local integrations 👌
April 8, 2025 at 11:03 AM
Thanks @timkellogg.me 🙏 - MCP truly shines in these local integrations 👌
/cc @timkellogg.me - more MCP stuff for you 😅
April 8, 2025 at 10:33 AM
/cc @timkellogg.me - more MCP stuff for you 😅
This is MIT license. So, please check it out on Github (github.com/eschnou/gepe...) and ping me if any issues or ideas for use cases to explore. Thanks!
GitHub - eschnou/gepetto: A minimalistic cli-based Operator
A minimalistic cli-based Operator. Contribute to eschnou/gepetto development by creating an account on GitHub.
github.com
April 8, 2025 at 10:33 AM
This is MIT license. So, please check it out on Github (github.com/eschnou/gepe...) and ping me if any issues or ideas for use cases to explore. Thanks!
I've tested it using open-source LLM. It works great with qwen-2.5-30b which is fantastic at tool use and smart enough to keep progressing in the task. Running it on a small instance with Ollama or a provider like @groq.com or @scaleway.com
April 8, 2025 at 10:33 AM
I've tested it using open-source LLM. It works great with qwen-2.5-30b which is fantastic at tool use and smart enough to keep progressing in the task. Running it on a small instance with Ollama or a provider like @groq.com or @scaleway.com
For browser operation, it uses Playwright in headless/accessibility mode. This allows for fast navigation and action. It doesn't support yet visual mode (which is heavier and requires multimodal models) but this is on the roadmap. Thanks @playwright.dev for the great work on the MCP connector 👌
April 8, 2025 at 10:33 AM
For browser operation, it uses Playwright in headless/accessibility mode. This allows for fast navigation and action. It doesn't support yet visual mode (which is heavier and requires multimodal models) but this is on the roadmap. Thanks @playwright.dev for the great work on the MCP connector 👌
At the core, this is a Agentic loop (chain-of-thoutght/ReAct/Tool use) with MCP integration. Written in Java, using Spring AI and Java MCP (thanks @tzolov.bsky.social and @thomasvitale.com 🙏)
Highly extensible, works with all popular commercial and open-source LLMs
Highly extensible, works with all popular commercial and open-source LLMs
April 8, 2025 at 10:33 AM
At the core, this is a Agentic loop (chain-of-thoutght/ReAct/Tool use) with MCP integration. Written in Java, using Spring AI and Java MCP (thanks @tzolov.bsky.social and @thomasvitale.com 🙏)
Highly extensible, works with all popular commercial and open-source LLMs
Highly extensible, works with all popular commercial and open-source LLMs
N'est ce pas plutôt Israel == USA et dans la situation actuelle on a pas intérêt à les énerver encore plus ? La situation est catastrophique.
April 7, 2025 at 10:27 AM
N'est ce pas plutôt Israel == USA et dans la situation actuelle on a pas intérêt à les énerver encore plus ? La situation est catastrophique.
Of course... je suis nostalgique d'une époque où l'on avait ce genre de discussions en direct, en publique, avec un membre de cabinet de ministre... sur ce bon vieux twitter 🤷
March 25, 2025 at 10:41 AM
Of course... je suis nostalgique d'une époque où l'on avait ce genre de discussions en direct, en publique, avec un membre de cabinet de ministre... sur ce bon vieux twitter 🤷
Athénée ou libre ? Si c'est un Athénée, tu peux m'envoyer la grille en MP ? Clairement ici plus possible et ils renvoient à la FWB.
Une grille où ceux qui ont prépa ingé n'ont pas de L2, et ceux qui ont prépa médecine ont math faible, je reste perplexe...
Une grille où ceux qui ont prépa ingé n'ont pas de L2, et ceux qui ont prépa médecine ont math faible, je reste perplexe...
March 25, 2025 at 10:40 AM
Athénée ou libre ? Si c'est un Athénée, tu peux m'envoyer la grille en MP ? Clairement ici plus possible et ils renvoient à la FWB.
Une grille où ceux qui ont prépa ingé n'ont pas de L2, et ceux qui ont prépa médecine ont math faible, je reste perplexe...
Une grille où ceux qui ont prépa ingé n'ont pas de L2, et ceux qui ont prépa médecine ont math faible, je reste perplexe...
Ha ! Merci pour cette recherche. Je retrouve les même chiffres. Mais difficile de trouver une documentation de la marge de manoeuvre de l'école (e.g. 4 ou 5h de Français, 2 ou 3h de gym, etc..).
March 25, 2025 at 9:30 AM
Ha ! Merci pour cette recherche. Je retrouve les même chiffres. Mais difficile de trouver une documentation de la marge de manoeuvre de l'école (e.g. 4 ou 5h de Français, 2 ou 3h de gym, etc..).
Ici c'est un Athénée. Donc officiel. D'après l'école ils ne peuvent plus offrir l'ancienne grille suite aux contraintes FWB sur le volume horaire. Mais je ne trouve pas la source officielle de ces règles. Perdu dans un labyrinthe de directives 😅
March 25, 2025 at 9:28 AM
Ici c'est un Athénée. Donc officiel. D'après l'école ils ne peuvent plus offrir l'ancienne grille suite aux contraintes FWB sur le volume horaire. Mais je ne trouve pas la source officielle de ces règles. Perdu dans un labyrinthe de directives 😅
Je viens de trouver le rapport. Assez long mais semble bien documenté. A creuser.
www.anses.fr/fr/system/fi...
www.anses.fr/fr/system/fi...
www.anses.fr
March 25, 2025 at 7:29 AM
Je viens de trouver le rapport. Assez long mais semble bien documenté. A creuser.
www.anses.fr/fr/system/fi...
www.anses.fr/fr/system/fi...
C'est assez dingue comme chiffres, ils annoncent une dose journalière de 2mg/jour (personne 100kg). Je ne trouve pas le moindre article/étude pour valider ces données. Les guidelines FDA sont de 50mg. Toutes les études que j'ai trouvé (pubmed) vont vers ces chiffres.
March 25, 2025 at 7:25 AM
C'est assez dingue comme chiffres, ils annoncent une dose journalière de 2mg/jour (personne 100kg). Je ne trouve pas le moindre article/étude pour valider ces données. Les guidelines FDA sont de 50mg. Toutes les études que j'ai trouvé (pubmed) vont vers ces chiffres.