Leonardo Gonzalez

VP of AI Center of Excellence, Trilogy |
AI Research & Development | Technical Leadership

[Leonardo Gonzalez @archimagos]

@Xinanexa @doodlestein @satyanadella Skills have an LLM operating on them at the top level, but a big advantage of skills in the first place is reducing this variability through consistency. Referenced assets, scripts, query files, etc. reduce ad-hoc coding & reasoning and add deterministic processing to the loop.

[Leonardo Gonzalez @archimagos]

@SemiAnalysis_ Ohhhhy I get it
Cause 7/2 = 3.5

I guess 🤷🏻‍♂️

[Leonardo Gonzalez @archimagos]

@0xSero @cherry_mx_reds Why not just define the models in ~/.codex/config.toml without needing to run a proxy?

[Leonardo Gonzalez @archimagos]

@atmoio It's shocking and at the same time not surprising at all
https://t.co/LrvKVLrCDw

[alphaXiv @askalphaxiv]

As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development

"Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning."

Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing.

This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider.

That is not safety. Safety policies should be transparent, auditable, and user-visible.

On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.

[Leonardo Gonzalez @archimagos]

@reach_vb I'm working on this to give codex superpowers for analyzing, planning, creating and editing videos:
https://t.co/5qLniYgzK3

[Leonardo Gonzalez @archimagos]

@danhdox @enjojoyy I'm building some primitives you could find useful for this:
https://t.co/5qLniYgzK3

[Leonardo Gonzalez @archimagos]

@DennisonBertram https://t.co/0vzwEihe3T

[Leonardo Gonzalez @archimagos]

@thsottiaux I would love to get your insight and engagement here:
https://t.co/TdBWa46Hkr

[Leonardo Gonzalez @archimagos]

@thsottiaux I've been loving the taste and the speed, but this is concerning:

https://t.co/VYMdSgBErK

The issue, and the meta-issue of how it was discussed and handled, is a departure from an excellent track record.

PLEASE let this be an anomaly and not a harbinger 🙏

[Leonardo Gonzalez @archimagos]

@benhylak @astuyve Could either of these support OpenHands? I'd love to explore integration options for OpenSymphony using openhands agent-server and soon codex app-server ...and of course every harness has different tool names 😆

[Leonardo Gonzalez @archimagos]

@fchollet You think people will still want to type?

[Leonardo Gonzalez @archimagos]

@feross Is there some universe where Socket gets compromised and malicious versions of packages get injected into all of their customers' environments?

Same reason I've resisted third-party password managers. Great reduction of attack surface but external single point of failure.

[Leonardo Gonzalez @archimagos]

@juliarturc I wonder why I haven't seen anyone talking about transformers having attention deficit disorder 🤣

[Leonardo Gonzalez @archimagos]

@petergyang If you hold the option key on a Mac (in Finder), the copy cmd becomes copy path

[Leonardo Gonzalez @archimagos]

@jxnlco I made this a year ago.
Could be time for a refresh...
https://t.co/7MRCV9KhX4

[Leonardo Gonzalez @archimagos]

@did0f I asked ChatGPT this a few weeks ago and it said that the selected model does not matter for automatic compaction, but it DOES matter for manual compaction.

[Leonardo Gonzalez @archimagos]

@AIandDesign @Meta This is how companies differentiate themselves these days

https://t.co/vBMjSQhiok

[Tuomas Artman @artman]

Today is a hard day. I shared this note with the @linear team today: We’ve made the difficult decision to increase our workforce. This is not a cost-cutting exercise or a reflection of anyone’s performance. We’re simply reimagining every role for the agentic AI era. We’re hiring. We’re sorry about that.

[Leonardo Gonzalez @archimagos]

@JackWoth98 @mattlam_ Thanks for everything you've done @JackWoth98
I hope to keep seeing good open contributions from you.

[Leonardo Gonzalez @archimagos]

@gregpr07 I've added this to OpenSymphony to allow debugging of current or past tasks, dropping into the OpenHands conversation.

I'll be adding codex soon 😃

[Leonardo Gonzalez @archimagos]

@theo How about their recent Prompt API steamrolling, too?