technologybriefs
8:48in productionCh. 1 · How it learns/ 8:48 · ceiling 15 min
AI · Products

ChatGPT

2022

ChatGPT is not intelligence — it’s a prompt-to-output relay, optimised for speed, not truth.

ChatGPT is a generative AI chatbot released by OpenAI on 30 November 2022. It uses generative pre-trained transformers (GPTs) to generate text, speech, and images. It was fine-tuned using supervised learning and reinforcement learning from human feedback (RLHF). It gained one million users in five days and 100 million in two months. In October 2023, DALL-E 3 was integrated for image generation. In March 2025, it was updated to use GPT Image instead. In February 2025, Deep Research launched — generating reports from web searches. In January 2025, Operator launched. In May 2025, Codex launched — capable of writing software, answering codebase questions, running tests, and proposing pull requests.

AIProductsArtificial intelligenceMachine learningProgrammingArtificial intelligence across the network →
Chapters & takeaways4
  1. 0:48
    How it learns

    It generates responses using GPT models fine-tuned with human-labeled examples and ranked outputs.

  2. 2:16
    Adoption velocity

    It gained one million users in five days — faster than any internet application before Threads.

  3. 3:43
    The image pipeline

    Image generation shifted from DALL-E 3 to GPT Image — a proprietary replacement with no public metrics.

  4. 4:58
    Agentic mode

    Deep Research and Codex turn ChatGPT into an agent — but with opaque latency, no source citations, and no audit trail.

Worth your time?

Yes. Study the whole thing.

3.5/ 5
What works
  • Rapid user adoption
  • Prompt-guided multimodal output (text, speech, images)
  • Functional agentic coding with Codex
  • Web-search-backed reporting via Deep Research
What does not
  • Reliably report verified facts
  • Explain its own reasoning process
  • Disclose operational cost or energy use
  • Provide source attribution in Deep Research
Study it if
  • Product managers prototyping conversational flows
  • Developers needing boilerplate code or test scaffolding
  • Designers iterating visual concepts via prompt
Skip it if
  • Researchers requiring citation-traceable outputs
  • Regulated industries needing audit trails
  • Teams lacking capacity to verify every output
The written brief1 min read

What it is and the problem it solves

ChatGPT is a generative AI chatbot released by OpenAI on 30 November 2022. It solves the problem of low-friction access to AI-generated text, speech, images, and code — not reasoning, but patterned output at scale.

How it works

ChatGPT uses generative pre-trained transformers (GPTs) to generate text, speech, and images in response to prompts. It was fine-tuned using supervised learning — where human trainers acted as both user and AI assistant — and reinforcement learning from human feedback (RLHF), ranking model outputs.

What works

User acquisition works: one million users in five days, 100 million in two months. Prompt-guided image generation via DALL-E 3 (October 2023) worked well enough to ship. Agentic features like Codex (May 2025) deliver functional software tasks — writing code, answering codebase questions, running tests, proposing pull requests.

What does not

It does not reliably report verified facts. Deep Research takes 3–30 minutes per report and depends on web search quality. RLHF introduces subjective bias without transparency. Image generation shifted from DALL-E 3 to GPT Image in March 2025 — a change with no stated performance or fidelity rationale.

What it changes

It changes how non-specialists access generative tools: bundling language, image, and code agents into one interface. It normalises agentic workflows — Codex proposes pull requests; Operator (January 2025) handles multi-step tasks — but without open benchmarks or runtime visibility.

Is it worth your time

Yes — if you need rapid prototyping of text, image generation, or code assistance, and can tolerate unreliability in factual reporting, latency in deep tasks, and opaque operational costs.

Same field · AI4 of 9
9:20
ELIZAELIZA demonstrates how thin the line is between pattern matching and perceived intelligence — and how readily humans fill the gap.
9:42
Fei-Fei LiImageNet is a foundational dataset, not an invention in the conventional sense — but its design, scale, and institutionalisation through ILSVRC redefined how progress in AI is measured, validated, and accelerated.
9:49
Geoffrey HintonGeoffrey Hinton is a researcher, not an invention. His impact lies in a series of specific, testable mechanisms—not a single breakthrough. He made neural networks legible as cognitive models. He did not deliver scalable production systems. His later proposals remain unproven in practice. His influence is structural, not infrastructural.
10:12
Google Search1997Google Search is a sequence of architectural decisions—not a single invention. Its value lies in operationalising scale, speed, and semantics. Its cost is opacity, drift from document to answer, and dependence on infrastructural choices that favour speed over auditability.
Up next in Technology

Commodore 64

· 9:35

Custom silicon—not marketing—made the C64 dominant. Everything else was cost engineering.

Play nextBack to AI
9:35