Makale
7708 keşif · Açık ağdan seçilmiş keşifler, en yeniden eskiye.
-
A Blank PDF Form Is My New Favorite AI Benchmark
GPT-5.6 impressed me with a 40-hour C++ implementation, while Grok 4.5 surprised me by reconstructing a fillable PDF. Together, they suggest better ways to judge models than one-shot answers.
bhart.org · EN
-
My AI Coding Setup Still Lives in the Terminal
WSL, herdr, and a custom tmux usage meter help me manage several coding agents without losing track of the work. The appeal is a workspace I can keep adapting.
bhart.org · EN
-
GPT-5.6 Found the Web Interface I Was Missing
Codex and GPT-5.6 helped me investigate a device I own and build a local browser interface after the official web option disappeared. It changed which personal projects feel worth trying.
bhart.org · EN
-
Fable Made the AI Compute Boom Make Sense
A few days with Claude Fable changed how I think about AI compute: capable local models could handle everyday work while large clusters make sustained investigations affordable.
bhart.org · EN
-
What Happens After AI Gets Smarter Than Every Human?
AI progress in mathematics raises a question I cannot shake: if machines eventually reason beyond our best minds, will their discoveries still make sense to us?
bhart.org · EN
-
fluffy rambles: My urbanist fantasy
I mostly love the town I live in , except that I really miss living in a more urban area with good walkability and transit.
busybee · EN
-
fluffy rambles: Technology extension and phone longevity
I’ve had the same phone for the past five years, an iPhone 13 Pro which I bought for two reasons:
busybee · EN
-
Coffee Talk: My coffee selection hierarchy
Recently, James Hoffmann conducted a coffee fermentation project , along similar lines to his previous decaf project . Since this project was based entirely around caffeinated coffee, and I am quite caffeine-sensitive, I opted to just do this as a risky “drink one cup every few days” thing rather than a formal cupping.
busybee · EN
-
fluffy rambles: Fatigue and support thereof
Over the past year or so my fatigue has been getting significantly worse. It’s almost always brought about by post-exertional malaise, which is one of the keystone symptoms of ME/CFS , which has no known treatment, just mitigations at best . And the core mitigation is to not overexert oneself.
busybee · EN
-
Notes: Men in tech #YesAllMen
busybee
-
fluffy rambles: ME/CFS + Fibromyalgia
At this point it is super clear that I probably have ME/CFS . This has not been determined by a doctor (not for lack of trying!), but by a lifetime of experience and the fact that the issues I have match the profile pretty much exactly. I’d be tempted to blame the bad case of COVID I got back in 2024 but the precursors were there for a long, long time, so if COVID did anything it was just to push me over an edge I had been approaching for a while.
busybee · EN
-
fluffy rambles: Accidental upgrade
So, a bunch of my services run, or rather, ran , on an old Intel NUC 8 that I bought back in 2020 and have mostly used as a “random stuff goes here” server. A year and a half ago the fan on it died, so I bought a cheap replacement fan on Amazon and all was good. And over time I started hosting more and more stuff on it, such as my owncast and PeerTube instances, and a few things for my home media streaming setup (which are also now accessible while traveling).
busybee · EN
-
Building Aligned Intelligence System. Part I: Creating GPT Assistant
In recent years, the field of natural language processing has witnessed a remarkable breakthrough with the advent of Large Language Models (LLMs). These models have demonstrated unprecedented performance in a wide range of language tasks, from text generation to question answering. However, the mathematical formulation of these models and the techniques used to fine-tune them for specific ta...
AstraBlog · EN
-
Building Aligned Intelligence System. Part II. Improving Large Language Models
In this post we will look at different techniques for steering LLMs behaviour to get desired outcomes, starting with some basic general principles such as writing a good prompt and ending with fine-tuning and augmenting models with external knowledge. Methods discussed in this post are mainly aimed at improving LLMs reliability and ensuring the consistency and factual accuracy of their...
AstraBlog · EN
-
Exploring Parallel Strategies with Jax
Training large language models either like GPT, LlaMa or Mixtral requires immense computational resources. With model sizes ballooning into the billions or sometimes even trillions of parameters, specialized parallelization techniques are essential to make training feasible. In this post, we’ll explore implementing some of these scaling strategies in Jax - a Python framework designed for hig...
AstraBlog · EN
-
Transformers Inference Optimization Toolset
Large Language Models are pushing the boundaries of artificial intelligence, but their immense size poses significant computational challenges. As these models grow, so does the need for smart optimization techniques to keep them running efficiently on modern hardware. In this post, we’ll explore key optimization strategies that are making LLMs faster and more memory-efficient. We’ll star...
AstraBlog · EN
-
Speculative Decoding: The Art of Guessing Well
Large Language Models spend most of their decoding time not computing, but waiting: every new token requires reading all of the model weights from memory, only to multiply them by a single vector. What if the model could check several tokens for the price of one? In this post we’ll derive speculative decoding from first principles, build a general framework that covers every modern drafte...
AstraBlog · EN