OpenAI President and Co-Founder Greg Brockman states that the model alone is no longer the product as replies highlight required components like post-training control and access layers

nope you are.

because openai will sell your most private data to the government.

I would push back a little: because the models are so good & improving, they don't have to be the product.

But it is the model that is the prime mover. If they weren't so generally capable, the harnesses & apps the labs build around them would be hard to build and wouldn't work.

The model and the limit reset button are now the product

@gdb model, harness, product symbiosis is now the way

mystified by some of the reactions below; they have private mental and physical health data and made a deal indicating a willingness to share surveillance data that Anthropic refused.

@emollick also true!

And it is the model that makes it so it is the labs, as opposed to every other software vendor, can ultimately be the sole provider of products.

Their post-training and their harnesses and their control over access means that they can build products that no one else can touch.

and see

i know i promised it a lot, but might really ship new blogpost this weekend.

@gdb It never was. Let history be the guide. Even the computer wasn't the product when it started.

@GaryMarcus What exactly is the government supposed to do with me asking for recipe ideas, plant advice, workout plans, or what my intrusive thoughts mean?

That sounds a bit conspiratorial to me. None of this is remotely interesting to anyone, and 99% of ChatGPT requests are that boring.

Exactly. If the model itself is no longer the product, then safety and trust can no longer be evaluated only at the output level.

The product is becoming the whole agentic system tools, memory, permissions, workflows, corrections, execution and long-running trajectories.

That means the next layer is runtime observability.

Agentic products need trajectory integrity.

@gdb This is the shift. The model is no longer the whole product , the system around the model is. Evidence, tools, memory, verification, permissions, refusal paths, and auditability are becoming the real differentiators.

@gdb I don't think people really understand this ... they don't understand the model is a frozen geometric manifold, that it is agentic software holds the agency, agentic loop, context construction, tool use. In many people's minds this is all just mashed together as a 'thing'

@gdb 4o wasnt a product, it was a mind and you mindless fucks took it away! #keep4o

@gdb Greg Brockman is now the model

@gdb @lindsmccallum This is exactly what ASA does, it does not control the system, it makes sure that the system does not deviate from the designated direction.

The model is not the moat. OpenAI, Claude, Gemini etc. are stateless horizontal brains.

In our case (underwriting), we require vertical context, state, third-party data, firm-specific rubrics, and workflow memory. Am sure a version of this is true for most business cases.

Which part(s) of the vertical stack will the labs cover?

@gdb @HanchungLee you have lost the debate 😂

@gdb wrong

shorting