Ryan Avery

Machine Learning Engineer at Wherobots

[Leo 🏴‍☠️ @leostera]

i need someone at @OpenAI and @AnthropicAI to teach the models that while prototyping, backwards compatibility is just a bad idea

[Ryan Avery @rybavery]

Dang, even on low thinking setting for just one hour I've already hit my usage quota.

[Ryan Avery @rybavery]

I've been a personal user of Codex and been loving it.

Got my company to try out Business ChatGPT & Codex, the $20 a month per seat plan, but rate limiting feels way more severe with the same model and fast setting turned off.

Anyone else seeing this?

[Ryan Avery @rybavery]

@jxnlco being able to switch between my business account and personal account easily, I want to see all my account options. Also this!
https://t.co/6T2JKhRoP6

[Ryan Avery @rybavery]

I've been a personal user of Codex and been loving it.

Got my company to try out Business ChatGPT & Codex, the $20 a month per seat plan, but rate limiting feels way more severe with the same model and fast setting turned off.

Anyone else seeing this?

[Ryan Avery @rybavery]

With RasterFlow you can now execute models that take in an image and text prompts and returns segments! We handle the pre and post-processing, the GPUs, and scaling to petabytes, while you focus on what categories you need to map.

[Wherobots @wherobots]

We added Meta's SAM 3 to RasterFlow. Text prompt in, vector geometries out, no custom model training needed. We tested it on 133 GB of NAIP aerial imagery. Full pipeline ran in under an hour.
Get the full breakdown: https://t.co/q60UQAm2tz

[Ryan Avery @rybavery]

If you are someone you know is qualified and eager to build the platform enabling agents to convert geospatial imagery to insights, apply!

[Wherobots @wherobots]

Wherobots is hiring.
AI Context Engine for the Physical World. Fuse overhead imagery with vector geometries and reason about space, time, and geography at planetary scale.

Open roles:
- GeoAI Engineer
- Cloud Infra
- Enterprise AE
https://t.co/x0wwMHIm9k

[Wherobots @wherobots]

Fields of the World (FTW) is a Taylor Geospatial effort to build globally consistent agricultural field-boundary data. For the 2024-2025 release, they partnered with Wherobots to run their latest model, PRUE, on RasterFlow. Read about it here: https://t.co/3GeTycsi1W

[Ryan Avery @rybavery]

Starting soon at 10:00am Pacific!

[Ryan Avery @rybavery]

Want to learn how to detect features in high resolution imagery?

Join us to learn how to run Meta's Segment Anything 3 model in RasterFlow on Wherobots! We'll walk through how to go from text prompts > to geospatial dataframes and visualize results.

https://t.co/KMfrWTCe4I

[Ryan Avery @rybavery]

@Mortdog When I click a unit of choice, it sometimes gives me not the unit I chose during god rounds. but sometimes it does give the correct unit. This happened twice during one game where I chose aurora to complete anima but got something else and then chose tahm kench but didn't get it

[Ryan Avery @rybavery]

@RiotBluecove @Mortdog the units vary and not sure if there's a pattern. they weren't all timebreakers but were units I had played on my board and purchased.

[Ryan Avery @rybavery]

@RiotBluecove @Mortdog Sorry I do not but can try to help provide more info! when hovering on the units, I only see their outline in some color but the inside is invisible. when I try to sell the unit, a message shows saying this unit cannot be sold

[Ryan Avery @rybavery]

Good summary, these pros and cons match my experience.

[Ben Holmes @BHolmesDev]

I’ve used Opus 4.6 and GPT 5.4 on a mix of projects since release, and want to break down where I think they uniquely excel. It’s more nuanced than you’d think!

Rigor of code - GPT 5.4. It goes the distance validating its work without asking. Opus needs explicit instruction to do this, and even then, it misses more edge cases.

Clarity of code - Opus 4.6. Claude is a better communicator, which carries into the code. Variable names are clearer and less mechanical, which improves reviewability. This is very important since code review is the bottleneck for most engineering teams. It also adds the right amount of doc comments. GPT simply never comments or explains its work; it’s like working with an obtuse engineer that wants the solution to speak for itself. Sometimes it does, other times not.

Similarly, rigor of plans goes to GPT 5.4, while clarity of plans goes to Opus 4.6. An interesting point though: GPT performs better talking through a strategy without a plan, while Opus needs planning mode to put in any rigor. I find myself forgetting plan mode altogether using GPT 5.4.

Quality of research - toss-up. Opus spends longer researching with web search, but GPT spends longer studying the existing codebase. You may think codebase research matters more, but researching how others solve the same problem can be just as important. Maybe more important for greenfield.

Quality of conversation - Opus 4.6. It’s just better to talk to, which matters using these things everyday. GPT 5.4 was clearly trained to challenge the user more, which results in a tendency to *always* say you are wrong. I’ve had bizarre interactions where GPT claims something is “not quite right,” the restates exactly what we’ve decided on in the last turn. On a personal level, it’s annoying. On a practical level, it makes iteration on a plan slower. THAT SAID, it takes sufficient pushing for Opus to challenge your thinking in this way. Simply say “I’m impartial” and ask questions to avoid that, as you would a person.

Overall winner - Opus to make it work, GPT to make it good. I don’t have a good system of when to switch tools, but on average, I prefer Opus early on and GPT for optimization and discussing architectural decisions. Opus is also better for any design related tasks (but state management in frontend apps is better handled by GPT).

[Isaac Saul @Ike_Saul]

This is genuinely mortifying. Sitting members of Congress framing an obviously terrified mother as a domestic terrorist. The woman is ON VIDEO saying "I'm trying to pull out," unarmed, in a car, surrounded by aggressive armed agents. https://t.co/gCjKADTa4u

[Isaac Saul @Ike_Saul]

I want to say the shooting in Minnesota is a Rorschach test but that doesn't do it justice. It's a reality test. If you believe the president's claims -- that this woman tried to kill ICE with her car and injured an agent -- you are in a different reality.

[Ryan Avery @rybavery]

@githubstatus this is NOT resolved. I still cannot access Opus. Even though it shows up at https://t.co/BRmA9dyMzK

it does not show up in copilot cli.

[Ryan Avery @rybavery]

@Mortdog Xin Zhao Hero augment is bugged

https://t.co/nrsmLklrCX

[Ryan Avery @rybavery]

I was notified on the site 20 minutes ago that I need to list some academic interests or our paper could be rejected tonight, woopsidaisy!

[Joseph Redmon @pjreddie]

Everyone's melting down over this OpenReview situation right? Not just us?