thatjoeoverthr

6 ms·

91.
by thatjoeoverthr 10mo ago
I actually had a fantastic experience with Fortran lately. I ported a compute kernel from python/numpy to Fortran 2018, partially due to the GIL and partly so I could use Intel's compiler. The performance improvement was tremendou
92.
by thatjoeoverthr 10mo ago
The analyses here miss the economic realities of building datacenters. "Just use land", "just use nuclear", "just use water". All of this is contested. A system of lawsuits and regulations turns negative extern
93.
by thatjoeoverthr 10mo ago
“Boo capitalism” on the outside, AI personhood on the inside. Actual agenda: disposable moral vehicles. AI liability is the goal. A tax on corporate profits is a tax on cost cutting already. A tax on “AI” is a way to compartmentalize. But y
94.
by thatjoeoverthr 11mo ago
All of this is true. “Invitation to guess” is the key phrase in my comment. CLIs present as cryptic, which is a UX _advantage_ over chat wrappers because the implied call to action is “go do some reading before you touch this”. An AI wrappe
95.
by thatjoeoverthr 11mo ago
In game design we used to call this opacity “hunt the verb” in text adventures. All chat bots suffer this flaw. GUIs solve it. CLIs could be said to have it, but there is no invitation to guess, and no one pretends you don’t need the manual
96.
by thatjoeoverthr 11mo ago
I used it exactly so (timers and alarm clocks) until after some updates it wouldn’t react to my voice even half the time. I haven’t tried it since in 10 years.
97.
by thatjoeoverthr 11mo ago
Maybe let's extend it to negative things, too. "We crashed a Yugo into a bollard."
98.
by thatjoeoverthr 11mo ago
I assume you're downvoted for pedantry (understandable) but it is a real pattern. Whenever it's a space topic it's always "we" or "Japan" or "America". Nobody is so vague on other topics. I suspe
99.
by thatjoeoverthr 11mo ago
Others have mentioned the large context window. This matters. But also important is embeddings. Tokens in a classic Markov chain are discrete surrogate keys. “Love”, for example, and “love” are two different tokens. As are “rage” and “fury”
100.
by thatjoeoverthr 11mo ago
Thank you for this work. It was my first C compiler in 1998. Y'all helped me on the mailing list and you even replied to my e-mails! I was 11 and this was insane to me.
101.
by thatjoeoverthr 1y ago
I should add. _Other_ children being on phones poaches them off, but currently enough are "let out" that it's not a total child desert. (This was harder for the previous child, long ago, so I think there is a trend.) A lot of
102.
by thatjoeoverthr 1y ago
Oh good question so there is a neighborhood, we’re surrounded by playgrounds and parks with other children and in the summer it’s surprisingly easy to keep busy. I did get her a bit of a kit (flip phone, radio, skates, nice bag) and she is
103.
by thatjoeoverthr 1y ago
I set up one with Ubuntu for my seven year old. 1. It's in the living room next to an SNES and an N64. 2. I showed her vscode, did a short HTML tutorial, and printed out some HTML cheat sheet. 3. Some modest games. 4. No YouTube or soc
104.
by thatjoeoverthr 1y ago
Honestly, just sad seeing AI posts on HN now.
105.
by thatjoeoverthr 1y ago
The Salesforce analyst quote is ChatGPT, though.
106.
by thatjoeoverthr 1y ago
I think no, categorically. The computer can detect your typos and accidents. But if you made a decision to word something a certain way, that _is_ your voice. If a second party overrides this decision, it's now deviating from your voic
107.
by thatjoeoverthr 1y ago
Even letting the LLM “clean it up” puts its voice on your text. In general, you don’t want its voice. The associations are LinkedIn, warnings from HR and affiliate marketing hustles. It’s the modern equivalent of “talking like a used car sa
108.
by thatjoeoverthr 1y ago
His agent hit what we in the biz call “max tokens”
109.
by thatjoeoverthr 1y ago
I've saved this and I'll study this when I come back to it. Thanks!
110.
by thatjoeoverthr 1y ago
Im just learning this from your text, after spending last week trying to get a BERT model to talk. https://joecooper.me/blog/crosstalk/ I’ve still got a few ideas to try though so I’m not done having fun with it.
111.
by thatjoeoverthr 1y ago
The thing with the hype is it's always the same hype. "If you can just 3D print another 3D printer ..." "Apps are dead, everything will be AJAX" etc. I no longer believe the hype itself warrants attention or pushbac
112.
by thatjoeoverthr 1y ago
For personal use, I won't tell Paramount :)
113.
by thatjoeoverthr 1y ago
Google has also a global weather model yielding by ten day predictions, and open street map runs local as well. Just today with GraphHopper and a map of Europe I can generate 2700 routes per second on my workstation. When I was young these
114.
by thatjoeoverthr 1y ago
There's a film about this, Multiplicity.
115.
by thatjoeoverthr 1y ago
I love it
116.
by thatjoeoverthr 1y ago
Same. Part of the old experience of building applications is the typing. You have some notion in your head of what you want, and now you've got to go _type_ for six hours to get some effect. This becomes a kind of "manual skill&qu
117.
by thatjoeoverthr 1y ago
There's good reasons to do this. Embedding similarity is _not_ a reliable method of determining relevance. I did some measurements and found you can't even really tell if two documents are "similar" or not. Here: https:
118.
RAG-Shot Learning(joecooper.me)
1 points by thatjoeoverthr 1y ago|0 comments
119.
by thatjoeoverthr 1y ago
I tried this with Instruct-3B now, and got the following text. "The company strongly advises against engaging in any activities that may be harmful to the environment.1` Note: The `1` at the end is a reference to the special text'
120.
by thatjoeoverthr 1y ago
I'm not prepared to run a larger model than 3.2-Instruct-1B, but I gave the following instructions: "Given a special text, please interpret its meaning in plain English." And included a primer tuned on 4096 samples, 3 epochs,

More ›