Borelli ConsultingArchiveSubscribe free
The Edge.
BY NICHOLAS BORELLIA clearer view of what’s next in AI.
Issue 008

04 / Resources

Resources

Nicholas Borelli · September 30, 2026

  1. Google: how language models workBackground on tokens and prediction.
  2. Google: training and weightsHow training changes the network’s numbers.
  3. GPT-3: Language Models are Few-Shot LearnersThe published 175-billion-parameter model.
  4. Scaling Laws for Neural Language ModelsModel size, data, and compute.
  5. Dario Amodei: We Must Pace the FrontierThe slowdown debate. This essay’s proposed approach is my own.
  6. Google DeepMind: AlphaFold · GraphCastFocused applications in molecular science and weather.
  7. Siemens: visual inspection · Predictive maintenance
  8. Meta: Introducing Muse · Zuckerberg interview · September 15 statementThe interview and later safety statement are separate sources.
  9. Harvey: Why Harvey Is Multi-Model by Design
  10. OpenAI: Introducing dots · Watch the launch videoSeptember 29, 2026. Product announcement and promotional demonstration.
  11. Meta: Llama 4 and the Behemoth previewNearly two trillion total parameters; model designs differ.
  12. IEA: Energy and AI · Epoch AI: training durationsData-center electricity use and the time required for training.
  13. Anthropic: recursive self-improvementThe possibility of AI accelerating AI development.

If the model you use stayed exactly as it is for the next year, what would you build with it? Tell me on LinkedIn.

— Nick

Plain English about AI for people running real businesses.

Written by Nicholas Borelli · Sources checked September 27, 2026 · Dots added September 29, 2026