Sent out my weekly-ish newsletter – this edition covers the latest prompt injections, gpt-oss, GitHub Codespaces, Gemma 3 270M, Qwen-Image-Edit and how congress. gov uses XSLT! https://
simonw.substack.com/p/prompt-injec
tions-as-far-as-the-eye
…
LLMS
-
Weekly AI digest: Prompt injections, open-source models, and development tools
By
–
-
Comparing Gemini and ChatGPT beyond simple chatbot classification
By
–
People keep comparing Gemini to ChatGPT like it’s just another chatbot.
-
Gemini demonstrating advanced strategic reasoning capabilities
By
–
Gemini is getting scary good at strategy level tasks. Not just summaries and rewrites anymore.
-
Reasoning Models Token Price Beyond Surface Analysis
By
–
De la gráfica anterior hay que tener en cuenta que hablamos del precio por token, y sabemos que los modelos razonadores exprimen sus capacidades generando más tokens… así que bueno, tomad la parte del precio como un indicador pero que no cuenta la historia completa.
-

AI Model Power Increases While Costs Decrease Year-Over-Year
By
–
Por visualizar el cuadrante inteligencia vs precio, el punto que marco es el modelo más potente que teníamos hace justo un año. Todos los demás, más baratos, más potentes… es lo que se ha logrado en los último 12 meses.
-
Modern SOTA LLM Reasoning Requires Ed’s Contribution
By
–
There's no modern sota LLM reasoning without you Ed
-
Open Source Performance Without Open Source Models
By
–
No open source model but open source perf. Sufficiently open ai I guess
-

OpenAI Plans Codex Code Review for PRs
By
–
OpenAI is working on a possibility to enable Code Review for all PRs via Codex.
-

Model-Driven FLOP Allocation Achieves 80% Sparsity, 4x CPU Speedups
By
–
What happens when you let a model decide how to allocate FLOPs? Introducing Compute Where It Counts, a new trainable sparsity paradigm released by @crystalAIorg that beats SOTA methods, enabling 80% sparsity and 4x+ speedups on CPU.
-
ElevenLabs v3 Text-to-Speech Model Now Available
By
–
ElevenLabs v3 is available through the “Speak” button below every bot response, as well as at https://
poe.com/ElevenLabs-v3 and across all platforms. You can also use this model through our new API. (2/2)