Unsloth Studio is more featureful (well integrated tool calling, web search, and code execution being headline features), and comes from the people consistently making some of the best GGUF quants of all popular models. It also is well documented, easy to setup, and also has good fine-tuning support.

Lately I’ve been playing with Unsloth Studio and think that’s probably a much better “give it to a beginner” default.

The trick is as others have said in what adding it to your design does in terms of complicating compliance design.

[0] https://www.digikey.com/en/products/detail/silvertel/AG9705-...

I (ironically, depending your penchant) used an AI judge to score every top HN post and its top comments all the way back to 2010 (over 300,000 individual judgments!), and created a nice interface with the historic charts.

The scoring is off of how much disgruntled AI pessimism the post and top comments show. You can use that score to either only show (in Doom Only mode) or hide (in calm mode) all the doom static.

This is part joke and partly to answer my own curiosity about how real my feeling that the comments and posts have grown increasingly overbearingly negative towards and focused on the worst of what results from AI in the past year.

The trends is stark. My intuition was reinforced by the analysis; disgruntled pessimism about AI is at an all time high in the last few weeks and on track to continue doubling annually since ChatGPT released.

Historic charts are at bottom of the page.

Comments URL: https://news.ycombinator.com/item?id=47608596

Points: 3

# Comments: 0

I (ironically, depending your penchant) used an AI judge to score every top HN post and its top comments all the way back to 2010 (over 300,000 individual judgments!), and created a nice interface with the historic charts.

The scoring is off of how needlessly "AI Doom" or AI curmudgeony the post and top comments are. You can use that score to either only show (in Doom Only mode) or hide (in calm mode) all the doom static.

This is part joke and partly to answer my own curiosity about how real my feeling that the comments and posts have gotten acceleratingly more overbearingly negative towards and focused on the worst of what results from AI in the past year.

I won't spoil the conclusion, you can have a look and judge for yourself now.

Comments URL: https://news.ycombinator.com/item?id=47601150

Points: 2

# Comments: 1

I wanted to test if this is still true: whether a "microgpt" in that spirit could be brought into existence with minimal manual intervention, just clear expression of intent to an LLM. This is an experiment not just in producing a tiny GPT artifact, but in seeing how close you can get to the essence of microgpt just through careful prompting, without writing a single line yourself.

Comments URL: https://news.ycombinator.com/item?id=47498833

Points: 3

# Comments: 1

The key insight was to not just handwave or guess at how much is automated, but make evaluation and review part of the continuous development loop. I first implemented in https://github.com/Entrpi/autoresearch-everywhere where I used it to deliberately automate more, in the spirit of Karpathy's upstream (and to very good effect. I have some of the best autoresearch results anywhere, and the platform is far more robust than it started).

0: https://entrpi.github.io/eemicrogpt/ (The writeup is from a few days ago, and I'm still running experiments before I do a big rewrite. Slowrun is good food for thought.)

0: https://entrpi.github.io/eemicrogpt/

In this toy regime, performance is set by granularity. When the work is a pile of tiny matrix multiplies and elementwise kernels, overhead and launch/scheduling costs can dominate peak throughput. Laptop CPUs can be faster than Blackwell GPUs. That’s a regime inversion: the “faster” machine can lose because it spends too much time on ceremony per step, while a simpler execution path spends a higher fraction of wall time doing useful math. In that corner of the world, a laptop CPU can beat a datacenter GPU for this workload - not because it’s a better chip, but because it’s spending less time dispatching and more time learning. That inversion reshapes the early-time Pareto frontier, loss versus wall-clock, where you’re trading model capacity against steps-per-second under a fixed time budget.

Early-time is where most iteration happens. It’s where you decide whether an idea is promising, where you map stability boundaries, where you learn which knobs matter and which are placebo. If you can push the frontier down and left in the first few seconds, you don’t just finish runs faster.. you change what you can notice. You turn “training” into feedback.

Inside, I take you on a tour of the AI engine room: how scalar autograd explodes into tens of thousands of tiny ops, how rewriting it as a handful of tight loops collapses overhead, how caches and SIMD lanes dictate what “fast” even means, why skipping useless work beats clever math, and how ISA-specific accelerators like Neon/SME2 shift the cost model again. The result is a ~19,000× speedup on a toy problem - not as a parlor trick, but as a microcosm of the same compounding process that drives real progress: better execution buys more experiments, more experiments buy better understanding, and better understanding buys better execution.

Comments URL: https://news.ycombinator.com/item?id=47240461

Points: 11

# Comments: 2

Detailed optimizing journey in the readme too.

Comments URL: https://news.ycombinator.com/item?id=47219623

Points: 2

# Comments: 1