iiinova — where your digital self keeps evolving

Alex Mirugwe

@mirugwealex.bsky.social

10 followers 9 following 10 posts

Computer Engineer (BSc.) | Data Scientist(MSc.) | Data Scientist | Al Researcher

Posts Replies Media

Alex Mirugwe @mirugwealex.bsky.social · 08/10/2026

Read Why medical AI can pass tests but still fail patients on Kudos: www.growkudos.com/publications...

growkudos.com

Why medical AI can pass tests but still fail patients

Artificial intelligence (AI) tools can perform well on carefully selected test data but still miss diseases or raise false alarms in everyday healthcare. This opinion article explores why that gap occ...

000

Reposted by Alex Mirugwe

Significance magazine @significancemag.bsky.social · 24/09/2026

While wealthier continents get excited about AI, many bright minds in Africa can only watch in frustration from the sidelines. Where is Africa in the AI conversation?, asks @mirugwealex.bsky.social academic.oup.com/jrssig/artic... #freetoread

011

Alex Mirugwe @mirugwealex.bsky.social · 11/08/2026

Anthropic is pushing AI provenance into the generation layer. Watermarked Claude text and C2PA image metadata offer signals for attribution and detection. The key test is robustness under paraphrasing, recompression, editing, and cross-platform distribution.

010

Alex Mirugwe @mirugwealex.bsky.social · 26/04/2026

Benchmark accuracy ≠ clinical value. A model can ace retrospective validation and still fail the patient. Nature Medicine calls for proportional evidence: stronger claims need stronger proof of real impact.

000

Alex Mirugwe @mirugwealex.bsky.social · 11/06/2025

I've seen too many companies rushing into AI out of fear of missing out. They're investing without clear use cases, no real understanding of ROI, and no plan for integration. Hype is not a strategy. This panic-driven approach will be costly.

010

Alex Mirugwe @mirugwealex.bsky.social · 08/06/2025

Apple’s new paper says reasoning LLMs don’t truly think. They generate reasoning-like text but fail on complex tasks. More tokens don’t help. Hallucinations happen when pattern matching fails, not because the model runs out of space but out of structure.

010

Alex Mirugwe @mirugwealex.bsky.social · 06/06/2025

Hugging Face and Google Colab just made model experimentation easier. You can now launch Colab notebooks directly from any model card. This update supports faster evaluation and better reproducibility in ML workflows.

010

Alex Mirugwe @mirugwealex.bsky.social · 30/04/2025

DeepSeek-Prover V2 just dropped - 89% on miniF2F - New SOTA on PutnamBench - Solves formal AIME problems - Uses RL to break math into subgoals Serious progress in formal reasoning Paper: github.com/deepseek-ai/...

010

Alex Mirugwe @mirugwealex.bsky.social · 26/04/2025

Deepseek R2 is almost here R2 is rumored to be a 1.2 trillion parameter model, double the size of R1 Training costs are still a fraction of GPT-4o Trained on 5.2 PB of data, expected to surpass most SOTA models Built without Nvidia chips, using FP16 precision on a Huawei cluster

000

Alex Mirugwe @mirugwealex.bsky.social · 24/04/2025

As models get bigger, they sound smarter. But that doesn't mean they're right. More data can mean more noise. Without careful filtering, large models just get better at making stuff up with confidence. Scaling needs curation, not just size.

000

Alex Mirugwe @mirugwealex.bsky.social · 17/11/2024

Check out my latest article: OpenAI's 5-Gigawatt Data Centers: A Technological Marvel or an Energy Crisis in the Making? linkedin.com/pulse/openais-… via @LinkedIn

linkedin.com

Discover thousands of collaborative articles on 2500+ skills

Discover 100 collaborative articles on domains such as Marketing, Public Administration, and Healthcare. Our expertly curated collection combines AI-generated content with insights and advice from ind...

000