Take Control Of Your Data. Join our Data Science Briefing newsletter for the best in #DataScience and #MachineLearning
Joined February 2015
- Tweets 7,404
- Following 461
- Followers 1,463
- Likes 351
1,713 Photos and videos
Pinned Tweet
Jun 12
"Just write a script" is often the right answer.
Until it isn't.
A thread on when LLMs beat traditional automation, and when they don't 🧵
1
96
Surpassing Frontier Performance with Fusion — OpenRouter Blog — Interesting case study on model fusion: a carefully chosen mix of cheaper models can beat top-tier single models on complex research tasks. Worth a… openrouter.ai/blog/announcem…
24
Jun 13
[2606.11058] The social consequences of AI delegation — Worth a look if you’re thinking about what happens when people hand off decisions to AI systems: it focuses on the social and organizational side effects, not just accuracy or efficiency.… arxiv.org/abs/2606.11058
18
Jun 12
Why does RAG hallucinate even when the answer is in your documents?
A thread on the 3 failure modes of vector RAG 🧵
1
82
Jun 12
Failure mode #3: Explainability
Vector RAG: "Here's a relevant chunk."
Graph RAG: "Here's the exact chain of entities and relations that led to this answer."
1
18
Jun 12
On July 11 I'm teaching how to build this from scratch — live, hands-on.
40% off → code BRUNO40
🔗 eventbrite.co.uk/e/productio…
19
Jun 12
"Just write a script" is often the right answer.
Until it isn't.
A thread on when LLMs beat traditional automation, and when they don't 🧵
1
96
Jun 12
The real unlock is structured output.
With Pydantic an LLM, you get:
👉 Parse messy input
👉 Return a typed Python object
👉 Plug it into any downstream system
Suddenly "unpredictable LLM output" becomes a non-issue in production.
1
41
Jun 12
I'm teaching exactly this in a live workshop on July 8th.
We build extraction pipelines, RAG systems, and multi-agent workflows with LangChain CrewAI.
If you write Python and want to go deeper: oreilly.com/live-events/auto…
29
Jun 11
Why Thermodynamics Rules Future Orbital Data Centers - IEEE Spectrum — A clear look at how basic thermodynamics—not just launch costs—will dictate whether orbital data centers make sense. Useful framing on cooling, power, and… spectrum.ieee.org/orbital-da…
21
Jun 11
Just announced: I'm teaching a live hands-on Graph RAG workshop on July 11.
You'll build a knowledge graph chatbot from scratch: raw text → entity extraction → graph → grounded LLM answers.
Tools: spaCy, REBEL, fastcoref, NetworkX.
1
49
Jun 11
Teaching a live 4-hour workshop July 8: build real LLM automations with LangChain, CrewAI, RAG, and Gemini. Hands-on, we write code together. For Python devs ready to go beyond the chatbot demo.
👉 oreilly.com/live-events/auto…
44
Jun 11
The 90-year-old idea behind JEPA models: Canonical Correlation Analysis (CCA) – Shon Czinner’s Blog — Useful refresher on how Hotelling’s 1936 Canonical Correlation Analysis underpins modern embedding-prediction… shonczinner.github.io/posts/…
22
Jun 10
No, Artificial Intelligence Is Not Conscious - The Atlantic — A clear, grounded argument for why today’s AI doesn’t qualify as conscious—and why stretching the definition leads to… theatlantic.com/philosophy/2…
21
Jun 10
The Bitter Lesson — A clear, practical reminder that general methods that scale with compute tend to beat hand-crafted tricks over time. Worth reading if you’re making choices between clever domain-specific hacks and… incompleteideas.net/IncIdeas…
25
Accelerating scientific discovery with AI — A clear look at where AI can genuinely speed up scientific discovery, and where the bottlenecks still are. Worth a watch if you want practical takeaways rather than broad… youtube.com/watch?v=yxAJohm0…
13
Estimating Infectious Disease Importation Risk during the 2026 FIFA World Cup | medRxiv — Useful preprint on how to estimate infectious disease importation risk during the 2026 World Cup—helpful framing for public… medrxiv.org/content/10.64898…
15