Brad Dwyer (@braddwyer) header
Brad Dwyer (@braddwyer) profile picture

Co-founder, CTO @roboflow: Give your software the sense of sight; use computer vision without becoming a machine learning expert.

San Francisco
Joined March 2009
Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)Photo by Brad Dwyer (@braddwyer)
Just submitted our 1st #madewitharkit app for iOS11! @magicsudokuapp solves Sudokus using #CoreML, #Vision & #ARKit. magicsudoku.com
110
1,878
529
4,003
Carbon is now available across @roboflow from Auto-Label, to Workflows, to Playground.
Gemini 4 Carbon is so fucking good at computer vision - #1 overall - #2 object detection - #1 short string extraction - #4 structured JSON extraction - #1 long text transcription - #1 text localization and recognition very slow and expensive though... ↓ examples
2
8
461
Founders, internalize this. Especially all our aspiring competitors
Replying to @MartinGTobias
If a company isn’t going from zero to 1 billion in under a month, they simply aren’t going to make it and should give up.
7
798
Argon finally aces this eval. Astra mixed up the M's and W's. playground.roboflow.com/shar…
Anyone have a model that can count the letters in this photo? All I have tried failed.
1
1
7
672
Argon is now the default model for labeling images in @roboflow, displacing Astra! Try it out in Playground, Workflows, Agent, and Auto-Label.
Astra dethroned Gemini 4 Argon is the best "vision" model #1 for detection: - tight, precise boxes - dense scenes #1 for text extraction: - small text - handwritten text - complex tables - rotated text 4× cheaper than Astra at high
2
1
23
1,776
A sufficiently large Zoom meeting is like a bag of microwave popcorn: the meeting is ready to begin when there's 3 seconds between the joining doorbell sounds.
6
463
We’re running 6.1 Sol on a portion of our traffic and still seeing 6.4% of conversations (1.9% of messages) fail with an error compared to 1.1% (and 0.2%) on 6 Sol. Significantly worse overall success metrics on that cohort as a result even though it does better on our evals.
Global reset landing tomorrow 10am PST for all paid ChatGPT accounts. Apologies for the slow start with GPT-6.1 Sol, it's now back to running at expected speeds after the massive load spike in the first two days.
5
385
Images labeled in @roboflow’s Auto-Label feature were up another 70% yesterday over last Thursday. We were able to cut costs 10x while keeping accuracy almost the same by making some prompt optimizations and switching the default to Sol 6.1 from Astra.
Almost a $2M run-rate yesterday. 😳
1
2
16
816
Only the most important checks happening in our CI.
4
16
864
Today I /goal’d Astra to speed up the loading time of our app. It worked for 13 hours and opened 63 PRs that, upon inspection, turned out to not improve the loading speed at all. These models are so hit and miss.
Ok this is pretty insane. Astra side-loaded a custom app for my DJI RC2 controller that takes a live video stream from my Mini 5 Pro drone, streams it to a @roboflow Workflow running RF-DETR via our Serverless Video API, & overlays the predictions back on the controller’s screen.
4
1
1
19
2,438
Almost a $2M run-rate yesterday. 😳
Roboflow users are already on a >$1M/yr run-rate for Astra auto-label consumption one week in. It seems the people like it and are getting great results!
1
1
1
11
1,785
Ok this is pretty insane. Astra side-loaded a custom app for my DJI RC2 controller that takes a live video stream from my Mini 5 Pro drone, streams it to a @roboflow Workflow running RF-DETR via our Serverless Video API, & overlays the predictions back on the controller’s screen.
41
53
11
771
61,099
I can run any model we support in Workflows. Qwen, Cosmos, SAM 3, you name it. Or custom RF-DETR models trained on things like roof damage, solar panels, or bridge cracks. So many cool demos incoming. Anyone want a tutorial?
13
72
2,093
The anatomy of a spear phishing attack; let me show you how this scam works.
This Post is from a suspended account. Learn more
2
2
396
A seemingly legitimate reporter (in this case, @misterbosquez) asks you to DM them about an article they’re writing related to your recent X posts. A couple of red flags here: 1, there is no reporter at Bloomberg with that and 2, my DMs are open.
1
324
Roboflow users are already on a >$1M/yr run-rate for Astra auto-label consumption one week in. It seems the people like it and are getting great results!
Astra is now the default model for auto-labeling in @roboflow.
1
1
26
5,136
The way to think about Astra is like a human; its job isn’t to run the same rote task over and over at scale, it’s to study and understand the task to automate it. Software engineers write code to perform tasks billions of times vs doing those tasks billions of times themselves.
I love Astra BUT I genuinely get ick whenever people use these models to tokenmaxx on some task they could solve with like 5k tokens using existing CV stack like @huggingface transformers or @roboflow under 200 LoC and low latency (real time)
7
6
22
4,254
Frontier models finally getting good multimodal cognitive understanding is _great_ for vision use-cases! Not because they’ll run in the hot-loop but because they’ll drastically lower the investment needed to go from “I have a camera” to reaping useful, ongoing business value.
1
3
181
Astra is now the default model for auto-labeling in @roboflow.
data labeling is dead GPT-6 Astra easily tells Celtics players from Knicks players whether they play at home or on the road
3
9
2
91
26,348
Capitalism is undefeated.
> "what's the easiest way to fix our comms issues?" > "maybe we should just cure cancer"
1
10
1,013
Using dumb models is like hiring dumb employees. Penny wise, pound foolish.
Claude Sonnet 5 is now on DeepSWE It scores below Opus 4.8, costs twice as much, and is even more expensive than Fable 5 Probably Anthropic’s worst release yet
3
11
693
Pro tip: use them to find ways to pay for their own token usage.
We asked Fable to check our cloud billing export against our codebase to find cost savings... and it found a place we were accidentally spamming logs to the tune of over $200k/yr. 🤯 Then it made a webpage touting its finds, including a video presentation of the highlights.
1
3
228
We asked Fable to check our cloud billing export against our codebase to find cost savings... and it found a place we were accidentally spamming logs to the tune of over $200k/yr. 🤯 Then it made a webpage touting its finds, including a video presentation of the highlights.
3
1
11
717