A double-week catch-up: gym streaks, a new router, and a chaotic Bellevue trip via LA featuring five-hour delays, missed connections, and Mr Murphy at every turn. Read more ›
A UK employment tribunal has ruled that Rockstar Games must face 'blacklisting' allegations in an upcoming trial brought by former employees. Read more ›
I picked up Ilustrado by Miguel Syjuco for the same reason I pick up most literary fiction I know nothing about — it won a pile of awards and the setting intrigued me. A murder mystery that opens in New York City and winds its way back to Manila? That sounded like exactly the kind of ambitious, globe-spanning novel I'd enjoy. Read more ›
GateMem: Benchmarking memory governance in multi-principal shared-memory agents. Read more ›
SIR KEIR Starmer has resigned as Prime Minister after months of turmoil and just two years in power. The tearful PM paid tribute to his "fantastic wife Vic, who who has been a rock by my side through good times and bad" as the couple emerged from Downing Street hand-in-hand. Victoria Starmer looked strained after... Read more ›
If travel demands freedom and spontaneity, parenthood would appear to require the opposite. Could this be why there seem to be surprisingly few fathers among the great travel writers of… The post appeared first on . Read more ›
Series — Fine-Tuning, Smallest to Largest: QLoRA (7B) ← you are here In , LoRA let me fine-tune a 1.5B model by freezing it and training tiny adapters. But the frozen base still sat in memory in 16-bit (~3GB). Now I wanted to go to Qwen2.5-7B — and hit a wall that LoRA alone doesn't solve. The problem A 7B model is ~15GB in 16-bit precision. A free-tier T4 GPU has 16GB. It would barely load, with no room left to actually train. The QLoRA insight QLoRA asks the question that naturally follows ... Read more ›
Open source skills file to ground AI assistants for legal, regulatory & compliance research. - Flipsta/Giblegal Read more ›
Last weekend I published the 100th episode of The Cycling Europe Podcast. If you listened to it, you will already know all about this exciting news; the audiobook rights for my four books have been purchased by a publishing company and, all being well, they will be available to streama and download later in the year. I've been asked to record the books myself (Stephen Fry wasn't available...) so I'll be spending quite a few days in an audio studio at some point over the summer doing just that. Read more ›
In its race to lead critical emerging technologies, China is working to expand adoption of its AI models and standards in the Global South. The Wire China looks at three very different countries in Southeast Asia to assess its progress. The post appeared first on . Read more ›
Defendants worry that changes could remove chance of acquittal based on jurors’ consciences in defiance of the lawClimate activists fear that delays to their cases may mean they lose the right to a trial before jurors, who are typically more likely to acquit them than a judge.Scores of defendants facing trials for protests as long ago as 2021 have had proceedings repeatedly postponed and worry that by the time their cases are heard, government changes limiting the right to jury trial may be i... Read more ›
Amit Chaudhuri at The Point: In the late 1990s, I read a short story by Nirmal Verma called “Terminal.” It had been written, like all of Verma’s fiction, in Hindi, and translated into English by the critic Alok Bhalla. I knew something of Verma’s work, because his reputation was a national one. I had taken… Read more ›
Back in 2022 and 2023 there were two big branches of machine learning happening at Meta1. The LLM work that led to Llama was a clean, smooth stack of repeated Transformer modules; the recommendatio… Read more ›
Two women have accused the former DUP leader of sexual offences. Read more ›
The Correspondent by Virginia Evans became a word-of-mouth bestseller last year and recently won the Women’s Prize for Fiction. It is an epistolary novel told entirely through letters, most of which are written over several years by septuagenarian Sybil van … → Read more ›
Natural language autoencoders are a really cool mostly-unsupervised method for producing free-form text explanations of LLM activations. You should read that paper (or the blog post) about them before reading this.I trained[1] several Qwen3-8B NLAs with different length penalties: during RL, I subtracted the token count multiplied by the length penalty hyperparameter (λ) from the RL reward[2]. I found that with small length penalty (λ=0.002), you can reduce the length of NLA explanations by ~... Read more ›
A jury found Mohammed Azim guilty of murdering Lily Whitehouse. Read more ›
You’re known for your strong voice, and you’ve talked a lot about how you developed that voice. I’m curious how you maintain it. I’m asking as a writer who, upon reading some of my work last year, was like, “I feel like I’m adjusting my voice for the audience rather than letting them come to me for my voice.” Read more ›
Twelve models worth knowing in 2026, each with one standout strength. Read more ›