Or, maybe, use AI to analyze and integrate UX research, and card sorting results ;)

My preferred Ollama model is `granite4:tiny-h`. It is fast, even on my PC without a GPU and with 32 GB of RAM. I use it for quick, easy questions. When the task is more difficult, I use GLM 5. I use proprietary, more expensive models only when I need to *escalate* because GLM fails. 1/3

I began debugging manually, but I realized the effort wasn't worth it. I therefore decided to use GLM 5 (via API), and it returned a working solution on the first attempt. 2/3

Today, I asked `qwen3.5:9b`, running via Ollama, to generate a Python script that takes the path of a SQLite database and a text file, writing the schema of the database tables to the file. The model failed to provide the correct solution. 1/3

Perplexity Perplexity is a free AI-powered answer engine that provides accurate, trusted, and real-time answers to any question.

Based on information from perplexity.ai/search/i-m-u..., I discovered how to toggle the "thinking" capability of a Large Language Model (LLM) running on Ollama.

You can use the `/set` command:

* `/set think` to enable thinking mode
* `/set nothink` to disable it
#til

Of VC-backed products, 78% fail:

42% fail because nobody wants them.
17% because they don't really solve the problem they claim to solve.
19% because they were no different from competing products or couldn't evolve with the market.

And that's just VC-backed companies.
1/2

Once upon a time, UI designers produced wireframes. Today we create #vibeframes.

Calvino, de Chirico. Italian artists were were quite ahead of their time. 😉

my competitive advantage is that i'm having fun

nvidia/parakeet-tdt-0.6b-v3 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

*Notes:* CPU-only inference is slower than GPU. I haven't benchmarked it against Whisper yet.
Model: huggingface.co/nvidia/parak...

A python script to run nvidia/parakeet-tdt-0.6b-v3 A python script to run nvidia/parakeet-tdt-0.6b-v3 - parakeet_cpu.py

I experimented with **nvidia/parakeet-tdt-0.6b-v3** — a multilingual ASR model — and was able to run it *on my laptop without a GPU*. I used local Python tooling and a tiny script (published here: gist.github.com/bussolon/aeb...). Tested with Italian speech and it transcribed correctly.

I would say pre-training is evolution, fine-tuning is childhood.

Create a map of Italy where every region is made out of its most famous food (the regions should actually look like they are made of the food, not a picture of the food). Check carefully to make sure each region is right.

Italians eat better ;)

G3mini?

Look closer, though, and a different story emerges. The race to power Al is creating a surge in electricity demand that clean energy is best equipped to meet. Data centers are signing long-term contracts for solar, wind, and battery projects, pulling private capital into renewables at a pace policy alone could never achieve. Because these technologies are cheap and fast to build, each wave of deployment lowers costs and accelerates the next, creating a powerful feedback loop between falling prices and rising capacity. Fossil fuel plants, which take years to permit and construct, simply cannot scale fast enough to keep up. As renewables expand to meet Al's appetite for power, their growing scale drives prices down further, eroding the economics of fossil generation and speeding the broader energy transition.

The AI Power Boom Is Accelerating the Energy Transition substack.com/inbox/post/175… #AI #renewables #energy #climate (good news)

This new AI technique creates ‘digital twin’ consumers, and it could kill the traditional survey industry A new paper details a revolutionary method allowing LLMs to accurately simulate human consumer behavior, poised to reshape market research. This technique creates synthetic consumers providing realistic product ratings and detailed reasoning. Existing AI market research struggles with LLMs producing unrealistic numerical ratings. The Semantic Similarity Rating (SSR) method prompts LLMs for textual opinions, converting them into numerical vectors. SSR's model achieved near-human accuracy when tested with a real-world dataset, mirroring human rating distributions. This development arrives as traditional survey integrity wanes due to AI-generated responses. This research offers a controlled way to generate high-fidelity synthetic data, shifting from data defense to offense. The method's success relies on quality text embeddings, accurately capturing purchase intent. This moves from analyzing existing data to generating novel insights before product launches. Digital twins of consumer segments can test product concepts offering rapid innovation cycles.

This new AI technique creates ‘digital twin’ consumers, and it could kill the traditional survey industry

A new paper details a revolutionary method allowing LLMs to accurately simulate human consumer behavior, poised to reshape market research. This technique creates synthetic con…

#ai #llm #news

Generate the image of a medieval manuscript showing the system prompt

Research on more than 600,000 college graduates suggests studying philosophy does make people better thinkers:

buff.ly/Re5yGyK

On the Theoretical Limitations of Embedding-Based Retrieval | alphaXiv View recent discussion. Abstract: Vector embeddings have been tasked with an ever-increasing set of retrieval tasks over the years, with a nascent rise in using them for reasoning, instruction-followi...

Limits of vector search

a new GDM paper shows that embeddings can’t represent combinations of concepts well

e.g. Dave likes blue trucks AND Ford trucks

even k=2 sub-predicates make SOTA embedding models fall apart

www.alphaxiv.org/pdf/2508.21038

Can't turn off the screen before bed?

30-min electronic free before bed improved performance for elite athletes (w/ bright light in a.m.):
www.ncbi.nlm.nih.gov/pubmed/40845249

Pro-atheletes do use blue-light blockers, destigmatizing these #psychiatry tools:
chrisaikenmd.com/darktherapy

#therapy

Spousal correlations for nine psychiatric disorders are consistent across cultures and persistent over generations - Nature Human Behaviour A study of millions of couples shows spouses often share psychiatric disorders. This pattern is consistent across cultures (Taiwan and Nordic countries) and has persisted for generations, impacting ho...

🧠 Couples share mental health risks

A study of over 6 million spousal pairs in Taiwan, Denmark, and Sweden found consistent correlations across nine psychiatric disorders, stable across cultures and generations.

🔗 www.nature.com/articles/s41...

#SciComm #MentalHealth 🧪

How can an imitative model like an LLM outperform the experts it is trained on? Our new COLM paper outlines three types of transcendence and shows that each one relies on a different aspect of data diversity. arxiv.org/abs/2508.17669

a trivia fact about this paper is that we submitted it to arxiv weeks ago, and it was hanging there in limbo for quite a while. apparently because we submitted to "AI" while they moved it to "HCI".

New open access paper in the Journal of Computational Social Science with mathematicians Louis Abraham and Charles Arnal:

"Prompt Selection Matters: Enhancing Text Annotations for Social Sciences with Large Language Models"

doi.org/10.1007/s420...

Today, we're breaking down the MIT study claiming 95% of generative Al pilots at companies are failing - and why this headline is misleading the entire market. The report, based on just 52 interviews and 150 survey responses, has been cited as a reason for Al stock crashes, but the methodology is deeply flawed and the findings are being wildly misinterpreted. What the study actually reveals is that while individual employees are getting massive value from Al tools (90% use LLMs regularly vs only 40% of companies buying subscriptions), organizations are struggling with implementation - not because the technology doesn't work, but because of leadership buy-in, poor change management, and organizational dysfunction.

No, 95% of AI Pilots Aren't Failing - YouTube https://www.youtube.com/watch?v=5QzqyrnL010 #AI #misconceptions

Recently, I've been regularly using Z-AI GM-4.5 LLM via OpenRouterAI, and I’m really impressed: it's affordable and offers very helpful responses.

This AI in edu podcast is extremely fun, funny, thought-provoking. My 14-year-old made off with my phone and second earbud to keep listening even though he is generally anti-AI and the podcast is more enthusiastic (though also critical). myrobotteacher.bsky.social calearninglab.org/myrobotteach...

Open models have made a pretty comprehensive shift in the last few years from an afterthought and debate point over AI safety to an industry norm

We’ll increase the pace of our coverage so long as the pace of releases stays this high

Here's the latest roundup of great open models
buff.ly/iLgtEx3

The individual, relational and societal costs of striving to feel good

Review by Brett Q. Ford

Web: go.nature.com/44LVyN4
PDF: rdcu.be/exQoL

#psychscisky #psychology

Long-term exposure to outdoor air pollution linked to increased risk of dementia An analysis of studies incorporating data from almost 30 million people has highlighted the role that air pollution – including that coming from car exhaust

Long-term exposure to outdoor air pollution linked to increased risk of dementia www.cam.ac.uk/research/new...

#dementia #Alzheimers #environment #pollution