mark_l_watson

5 ms·

91.
by mark_l_watson 3mo ago
I used to do this to build command line utilities in SBCL, with millisecond initialization times. Good technique, and compress the image so executables are fairly small.
92.
by mark_l_watson 3mo ago
Clog is very cool. As an alternative, I have recently been building Common Lisp UIs with webview (I have two example webview apps as recent additions to my Common Lisp book). re: heap based delivery: not a good idea. But, I sometimes do hea
93.
by mark_l_watson 3mo ago
To me, the most weasel words are “”All sufficiently capable models, open and closed, should go through mandatory safety testing.”” I bet via regulatory capture that evaluating Chinese models will be a very slow process if Dario gets his way
94.
by mark_l_watson 3mo ago
Welcome to our political system. There are ways to fix this, in theory, but our government has passed the point of no return with corruption. There is one solution: no one register as a democrat or republican: this would make the dirty D.C.
95.
by mark_l_watson 3mo ago
I agree, then need to bring back Nelson Minar’s old search API, or something like it. Google does supply search grounding with Gemini API calls, and that is handy, but not general enough.
96.
by mark_l_watson 3mo ago
I more or less agree. When Apple starts using Chinese and expanded South Korean production memory chips, their hardware pricing problem will go away. I am running new betas for macOS/iOS/iPadOS and Siri is actually useful for a mu
97.
by mark_l_watson 3mo ago
Great show of principles and character, +1000 to the author. What can we "regular people" do? My approach, which is very inconvenient, is to try as much as possible to not feed my consumer spending money to tech giants. In some ca
98.
by mark_l_watson 3mo ago
That sounds reasonable, Intel Macs aren't much used anymore.
99.
by mark_l_watson 3mo ago
My problem with AI use of em-dashes is that it covers up sloppy writing. I simply collected about six examples of horrible use of em-dashes and how I would have written the text. This became part of my SKILL for general writing (along with
100.
by mark_l_watson 3mo ago
I am repeating myself: at least a partial solution is enforcing anti-dumping laws, that is, make money losing subscriptions illegal. If businesses and individuals pay actual token costs, then they will naturally only use tokens in ways that
101.
by mark_l_watson 3mo ago
This seems like a smart move, given their ability to host efficiently. I approve of efforts to make the cost of inference for smaller useful models slowly approach 'close to zero' and there are many good paths for getting there. I
102.
by mark_l_watson 3mo ago
This is a real problem, a long time in the making. Over fifty years ago advertising started to affect the way people thought: I argue that they trained people to not be content with a classic simple life. Fast forward to the best tech focus
103.
by mark_l_watson 3mo ago
Why? The current administration is doing such a fine job protecting the investments of a few hundred wealthy people, and not caring about the economy of the country. The combination of a wide range of model sizes and capabilities, with good
104.
by mark_l_watson 3mo ago
+1 I have read most of Rand's works. She grew up in the Soviet Union and hated collectivism and carried it very much too far. Still I find value in her work, and if she were here today I would bet $100 that she would disapprove of the
105.
by mark_l_watson 3mo ago
Great 'scratch my own itch' story and I like the framing of Emacs and Elisp as Malleable Software.
106.
by mark_l_watson 3mo ago
I share the cynicism here about this article. I think the saddest part of betting our economy on AI is that Google is actually the US-based corporation that is best positioned because so much of their AI efforts are not specifically aimed a
107.
by mark_l_watson 3mo ago
For me, poolside.ai “came out of nowhere” a week or so ago when I discovered their local coding harness ‘pool’ and their smaller 33G MOE model that runs fast and is effective on my old 32G mac mini. Really good work!! I need to evaluate the
108.
by mark_l_watson 3mo ago
I like that v4 flash is so fast! I run it on both FireWorks.ai in the US and bought some tokens directly from DeepSeek as an experiment. I only work on Open Source projects, so I don’t have to worry about my work being used to train models
109.
by mark_l_watson 3mo ago
There is a ton of headroom (or room for improvement) in smaller locally runnable models. Some of the Gemma 4 models were re-released this week with better tool support and the improvement in using it with pi for a local coding harness is ve
110.
by mark_l_watson 3mo ago
I found qwen3.6:26b slightly better on my 32G mac mini than the same sized gemma until gemma was updated with better tool support 4 or 5 days ago. It is like a ping-pong game: the advantage flips back and forth between providers.
111.
by mark_l_watson 3mo ago
You may be correct but the combination of local models when they are fast enough and work, combined with paying close to zero for deepseek v4 flash from US providers is pretty good. When you need it, glm5.2 is cheap to use and very good for
112.
by mark_l_watson 3mo ago
I worry about that but here are two positive things: 1. it would put us (USA) at a competitive disadvantage, and cooler heads will prevail in this fight 2. there are good US open models. I have the latest gemma4:27b with better tool support
113.
by mark_l_watson 3mo ago
Very true. The code is interesting and educational since the overall design is simple. Also true that the PDF to vector database code reminded me of something I would have done a couple of years ago. (I almost always now combine good old fa
114.
by mark_l_watson 3mo ago
Interesting looking project, but I just looked at the repo. UIUC is such a great school. I managed an AI team in the UIUC Research Park for two years and interactions with professors and hiring a lot of student interns was a good experience
115.
by mark_l_watson 3mo ago
Apple’s System Model is actually pretty good but is hard-limited to a 4K context length. This is OK for small Python utilities that need a model for applications that operate on small amounts of data, but is a disappointing limitation. That
116.
by mark_l_watson 3mo ago
re: > Right now unless you're paying by the token, there's no cost based reason to use the open weight models for daily coding work because the monthly coding plans from Anthropic and OpenAI are a better deal. Maybe. I am on a
117.
by mark_l_watson 3mo ago
That is a nice business niche, but I would hope that companies might also care about security. There are just a few inference providers who I trust to not save data outside the scope of maintaining inference cache, but I do use untrusted pr
118.
by mark_l_watson 3mo ago
I was blown away by how effective their latest model drop that works with their own coding harness pool is. I tested it extensively with Haskell, Python, and TypeScript for small coding projects. The functionality is good but the inference
119.
by mark_l_watson 3mo ago
+1 I enthusiastically use Chinese open weight models for a wide range of tasks (I also love Opus and Gemini) but I am so happy to see another high quality American open model (I consider gemma to be high quality, like qwen). I enjoyed turni
120.
by mark_l_watson 3mo ago
I retired several years ago and now I spend my allotted ‘tech time’ learning. Thirty five years ago I used to make up for having the ‘wrong’ degree (Physics instead of Computer Science) by writing books (Springer-Verlag, McGraw-Hill, etc.)

More ›