OpenAI rolls out DALL-E 3 access to ChatGPT Plus and Enterprise subscribers, after preparing a safety mitigation stack to ready the model for expanded release

all in real-time. https://openai.com/... [image] Greg Brockman / @gdb : DALL·E 3 is now available to all ChatGPT Plus & Enterprise users: @swyx : In a Surprising-for-these-times move, @OpenAI actually published a research paper detailing the improvements they made for DALL-E 3 - primarily “caption improvement” via a finetuned image captioner (GPT-V?) + upsampling descriptions with GPT-4 (prompt included).... [image] Forums: Hacker News : Improving image generation with better captions [pdf]

The Verge 2023-10-20 Umar Shakir

Context & Ripple Effects

DALL-E 3’s release follows OpenAI’s earlier plan to make the model callable from ChatGPT for paid and enterprise users, turning image generation into a conversational feature rather than a separate creative tool.

The rollout also lands after Microsoft exposed DALL-E 3 through Bing products, making OpenAI’s own subscription tiers a more direct channel for monetizing and governing access.

First-order effects

ChatGPT Plus and Enterprise subscribers can generate images through DALL-E 3 within ChatGPT, expanding the practical value of those paid tiers.
OpenAI’s safety mitigations and documented captioning improvements become part of the product’s release gate, tying broader access to managed deployment rather than model capability alone.

Second-order effects

Paid ChatGPT tiers become a clearer competitive bundle: rivals offering text assistants face pressure to integrate image creation and controls into the same workflow.
Enterprise customers gain another generative-media capability inside an existing interface, increasing the importance of account-level access controls and safety policies in vendor evaluations.

Third-order effects

If this staged pattern persists, multimodal features will increasingly be commercialized through subscription and enterprise access tiers, with safety systems functioning as core product infrastructure.
The move points toward conversational interfaces becoming the primary distribution layer for generative tools, concentrating both user workflow data and policy enforcement in assistant platforms.

The trend: Generative-AI vendors are packaging multimodal creation inside paid assistant products while using staged access and safeguards to control expansion.

Discussion

@main_horse @main_horse on x
sooooo is nobody talking about the research paper that got published with this post? [image]
@andriy_mulyar AndriyMulyar on x
DALLE-3 is trained on almost solely synthetic data produced by another AI model. Now that the internet has been squeezed dry - the future is training on other AI model outputs. Research report: https://cdn.openai.com/...
@mf_foom MF Foom on x
fascinating tidbit from the dalle-3 paper on where things might be going they suspect text rendering is degraded by the fact that their T5 text encoder is ~word-level ∴ exploring character-level text encoders might lead to better text rendering https://cdn.openai.com/... [image]
@openai @openai on x
DALL·E 3 is now available to all ChatGPT Plus & Enterprise users, letting you create unique images through conversation. Describe your vision, let ChatGPT generate multiple variants, and then request edits — all in real-time. https://openai.com/... [image]
@gdb Greg Brockman on x
DALL·E 3 is now available to all ChatGPT Plus & Enterprise users:
@swyx @swyx on x
In a Surprising-for-these-times move, @OpenAI actually published a research paper detailing the improvements they made for DALL-E 3 - primarily “caption improvement” via a finetuned image captioner (GPT-V?) + upsampling descriptions with GPT-4 (prompt included).... [image]

Chronicles