Joined March 2023
- Tweets 1,502
- Following 38
- Followers 37,827
- Likes 1,178
308 Photos and videos
Pinned Tweet
May 12
1/ Most AI systems can tell you what a model predicted. Very few can show exactly how it arrived there.
This is a look inside Sertn’s Proof Inspector: a 3D layer for verifiable inference.
From the original frame → activations → proof artifacts → final detections.
189
420
501
14,838
The future of AI won't be defined solely by larger models or faster inference.
It will be defined by systems that can operate reliably in the real world, where decisions carry consequences and trust matters.
That's the future we're building toward at Inference Labs.
inferencelabs.com/
212
111
263
2,746
Jun 12
We're building more than infrastructure. We're building a community around the future of verifiable AI.
Whether your interests are computer vision, AI systems, cryptography, deployment, or emerging technology, there's a place for you in the discussion.
Join our socials:
discord.com/invite/inference…
t.me/inference_labs
217
101
271
2,945
Jun 11
NVIDIA is increasingly positioning its next generation hardware around AI agents, not just AI models.
For years the conversation was: "How do we train better models?". Now it's: "How do we operate autonomous systems safely at scale?".
Very different challenge.
218
109
266
3,275
Jun 10
Anthropic's IPO filing is another sign that AI is becoming infrastructure, not experimentation.
As the industry matures, governance, auditability, and verification matter more.
The bigger AI gets, the harder "trust us" becomes as a strategy.
221
113
283
5,843
SpaceX, Anthropic, and OpenAI are all being discussed as potential IPO candidates.
That says something important about where the market is. Investors are no longer valuing AI purely on research breakthroughs. They want durable businesses, revenue, and production adoption.
45
38
98
2,157
1/ What is Sertn?
At its core, Sertn.ai is a computer vision platform built for real-world deployment.
Teams can annotate data, train models, deploy computer vision workflows, and monitor results from a single environment.
68
44
162
3,622
2/ But we also focus on something that's often missing from AI deployments: verification.
Instead of relying solely on logs, Sertn can attach a cryptographic proof to an inference, helping create a verifiable record of what happened.
As AI moves into operational environments, trust becomes part of the product.
9
743
Major publishers are continuing to challenge how AI models are trained.
Whether the courts ultimately side with AI developers or rights holders, one thing is becoming clear:
Provenance, transparency, and verification are moving from nice-to-have features to business requirements.
134
59
224
2,432
Everyone is talking about AI agents.
Far fewer people are talking about verification.
As AI systems move from copilots to autonomous workflows, the question shifts from "Can it do this?" to "Can we trust what it did?"
That's where the next wave of AI infrastructure will be built.
inferencelabs.com/
149
61
234
2,401
Inference Labs retweeted
Loss curves provide a really low bandwidth picture of what's really going on within models. With this visualizer, it becomes easy to understand what specifically needs to be done to improve the model.
1/ Training a model isn't just about accuracy.
It's about understanding 'what the model is actually learning' and where it is struggling.
Sertn's Model Art visualizes class performance in real time, helping teams identify underrepresented classes, missing samples, and training blind spots before they become deployment issues.
sertn.ai/
39
3
67
2,032
Computer vision is becoming easier to deploy.
Trusting the output is still hard.
At Sertn, we're building a workflow that brings annotation, training, deployment, and verifiable inference together in one platform, helping teams move from experimentation to production with confidence.
sertn.ai/
148
61
223
2,204
1/ Training a model isn't just about accuracy.
It's about understanding 'what the model is actually learning' and where it is struggling.
Sertn's Model Art visualizes class performance in real time, helping teams identify underrepresented classes, missing samples, and training blind spots before they become deployment issues.
sertn.ai/
141
74
231
4,301
2/ Instead of digging through logs, users can explore how individual classes evolve throughout training.
Rich class diagnostics make those insights accessible even to non-computer vision users, helping teams quickly understand model behavior and potential areas for improvement.
2
2
4
831
For years the industry focused on model training. Now we're watching billions of dollars flow into:
* data infrastructure
* deployment
* inference
* governance
* verification
119
80
235
3,084
SN2 Dsperse has now surpassed 1B proofs completed.
Proof throughput continues to accelerate, with peak volume reaching 19.4M proofs/day as larger models become increasingly practical to verify at scale.
Verifiable inference infrastructure is moving into production.
subnet2.inferencelabs.com/
146
79
318
2,604
May 31
1/ Regulators, auditors, and legal teams are all asking the same question in 2026: "Show me what happened and show me the control path behind it."
Logging outputs isn't enough. That's not an audit trail. That's just observing an outcome.
129
38
220
2,336
May 31
2/ The difference between output logging and runtime proof is where most enterprise AI governance breaks down.
Inference Labs makes inference itself the evidence.
Not a record of what happened. Proof that it couldn't have happened any other way.
1
2
15
924
May 30
1/ Over 40 bills introduced across 25 US states this year require clinical oversight of AI tools.
Utah now mandates disclosure when AI reviews authorization requests.
The question no law has answered yet: how do you verify which model actually ran?
143
39
212
2,187
May 30
2/ A hospital can license an FDA-cleared CV model and quietly serve a cheaper one.
No patient, no auditor, no regulator can tell the difference today.
Verifiable inference is how that changes. That's the problem Inference Labs exists to solve.
1
1
11
831