How to Read a Model Release Without Being Sold To
Every launch comes with a chart showing the new model winning. Here is what the chart leaves out, and which parts of the announcement are worth your attention.
AI News Round covers artificial intelligence — the breakthroughs, the policy battles, the business moves, and what it all means beyond Silicon Valley. Daily news. Weekly analysis. An African lens on a global story.
Every launch comes with a chart showing the new model winning. Here is what the chart leaves out, and which parts of the announcement are worth your attention.
The release notes list improvements across the board. Testing against them finds gains in two areas and no measurable change in most of the rest.
The demo works. The rollout does not. What sits between them is rarely the model, and companies keep being surprised by the same four things.
The first comprehensive AI law is now operating. Its risk-tiered structure is widely copied — including the parts that have not worked.
The useful unit of analysis is the task, not the occupation. Almost no job is fully automatable, and almost every job contains tasks that are.
Plant disease detection from a phone camera is a well-suited problem for machine learning. Getting it to work for a smallholder in the Far North is a different question.
Most tool round-ups ignore whether you can sign up, pay, or run the thing on the connection you actually have. This one starts there.
Research Digest: What the Chinchilla Paper Still Gets Right About Scaling Laws
A 2022 paper on the ratio of parameters to training data reshaped how models are built. Most of its conclusion...
Opinion: AI Governance in Africa Cannot Be an Afterthought
African governments are being invited to comment on AI frameworks drafted elsewhere. Consultation after the fa...
AI in Cameroonian Agriculture: Three Startups Trying to Solve a Real Problem
Plant disease detection from a phone camera is a well-suited problem for machine learning. Getting it to work...
What the Latest GPT-4o Update Actually Changed — And What It Didn't
The release notes list improvements across the board. Testing against them finds gains in two areas and no mea...
The Small Models Are the More Interesting Story
Frontier releases get the coverage. Models small enough to run on a laptop or a phone are changing more about...
Which AI Tools Are Actually Useful for African Developers? An Honest Assessment
Most tool round-ups ignore whether you can sign up, pay, or run the thing on the connection you actually have....
African Languages and AI: Why the Gap Is Bigger Than Most Reports Admit
Benchmarks that report African language coverage tend to measure the wrong thing. What breaks is not translati...
Why Most Corporate AI Pilots Never Reach Production
The demo works. The rollout does not. What sits between them is rarely the model, and companies keep being sur...
Plant disease detection from a phone camera is a well-suited problem for machine learning. Getting it to work for a smallholder in the Far North is a different question.
Benchmarks that report African language coverage tend to measure the wrong thing. What breaks is not translation quality but everything downstream of it.
Streaming responses, hopeful reconnects and generous payloads all assume a connection that a great many users do not have.
Perplexity AI
OpenAI
Midjourney
Hugging Face
ElevenLabs
GitHub / Microsoft
A 2022 paper on the ratio of parameters to training data reshaped how models are built. Most of its conclusions have held up.
African governments are being invited to comment on AI frameworks drafted elsewhere. Consultation after the fact is not participation.
Arguing about a hypothetical future system is more comfortable than arguing about the deployed ones. That is largely why it is so popular.