Interested in natural language processing, information retrieval/search engines, machine learning, software technology, GIS, start-ups, R&D, innovation, security, UNIX and ethical & social impact of technology.
AI research professor during the day, startup CTO at night, coder, avid reader and book collector.
Wrote my first compiler during high school. Systems I've had a hand in are used by folks ranging from London traders to the Justices of the U.S. Supreme court.
Submissions
Better email search/contact management?
1 points·by jll29··0 comments
BiasScanner: Detect 27 Types of Media Bias & Propaganda Using AI
biasscanner.org
4 points·by jll29··1 comments
Ross Anderson Has Died
therecord.media
2 points·by jll29··1 comments
[untitled]
1 points·by jll29··0 comments
Zuckerberg and Sandberg asked to provide testimony about Cambridge Analytica [pdf]
storage.courtlistener.com
7 points·by jll29··0 comments
“My GPT-3 Equipped Microwave tried to kill me”
youtube.com
1 points·by jll29··0 comments
Process Methodology for Data Science “An Evaluation-First Methodology”
arxiv.org
2 points·by jll29··0 comments
How to Resurrect Blackberry Devices?
3 points·by jll29··0 comments
Ask HN: Best low-cost, zero admin X terminal?
2 points·by jll29··0 comments
Re-Thinking Hacker News: From HN to Hacker Villages (HV)
I don't know who downvoted the parent or why, but it's a fair question IMHO.
The answer is there can be dramatic difference running a benchmark one time, because LLMs are not deterministic.
A proper methodology would ask each question 20 times and calculate the mean correctness across experiments.
The reason is that the temperature parameter introduces random behavior.
I understand you. As an AI research (actually, despite formally called professor of AI, I have historically always
called myself more specifically NLP/IR/ML researcher rather than "AI" researcher, to avoid historic baggage).
I am intellectually curious whether "intelligence" and "consciousness" are simply layers that can be separated from the hardware they operate on or first emerged from, but
I never had the urge or saw the point to practically replace
human beings by machines.
Not everything that can be done should be done.
When I go to a supermarket, I always avoid self check-out. Why? First, I enjoy human interaction, even if machines may be more efficient (not yet the case). Second, I don't want the people working at the checkout to lose their jobs.
When we purchased a GPU cluster, I felt bad when I learned that the electricity bill was going to be north of 60 000 EUR every year (that number apparently assumes no jobs are running - so 100% idle time).
Note that there is a certain level of arbitrariness
involved in this association game. For instance, if a household regularly is in need of both parsley and also condoms, the fact that they are purchased together may be a result of the pure coincidence that both were empty/used up at the same time (which is also a function of the package sizes of both items). We would be much less surprised at the mined associations if we took a longitudinal,
per-household look.
Furthermore, a shopping basked is per-household, but not per-person: the parsley and the condom may be used by different members of the household, or be shared, or be part of a gift to someone outside.
The human brain also tends to make up "causal" connections between any two items, when the real reason is often much more mundane.
There is a version of Chekhov's short stories available on Amazon [1] that is bi-lingual.
Has anyone tried this? Are the translations good? Is the material sentence-aligned so that it is easy to switch between reading the original and the translation?
Always a good exercise to implement a text editor, every programmer should have tried it. And I always try using new ones, curious to find something exciting.
Sorry, but it looks exactly like Zed.
And I'm not going to try out Zed again until they finally introduce auto-save, which Sublime has had for a long time and Emacs had since last century (because losing a file hurts you more than GPU rendering pleases you).
The title of the HN post "History of T (paulgraham.com)" is
isleading, as this article was not written by Paul Graham
himself. "Olin Shivers: History of T (paulgraham.com)" would
be clearer.
Thanks for the personal take on Scheme history.
> It ran on Vaxes & 68000's, which had also just come out.
It ran on Vaxen and 68000s, which had also just come out.
> All hot compilers do DFA.
Note here, DFA stands for Data Flow Analysis (not Deterministic
Finite Automata, another compiler acronym).
> the deepest and most powerful part of my diss, in my opinion, is the part (a) about which no one seems to know and (b) which is on the shakiest theoretical ground: environment reflow analysis. I would surely love it if some interested character one day takes that piece of my diss and really takes it someplace.
I know that feeling. But my experience is the author is best-suited to do that themselves.
Receiving one of Don's cheques ("Bank of San Serif" ;-) a few months after pointing out an error has been many a computer scientist's career highlight!
I agree with that opinion. He started writing TAOCP in 1968, and could have switched to Pascal in 1972.
Pascal is simple and clear, and can be translated easily to anything from LISP, Fortran, Python to C or C++ (in fact,
subsets of Pascal are often used as sample language in books
about compilers, including in Pascal inventor N. Wirth's own compiler book (which, unlike Knuth's, was completed timely):
It does not matter that Pascal is not much in use anymore, because due to its readability, it's timeless. It nearly reads like English prose, yet is automatically executable. It has also been standardized, and there is a book-sized language description available, as are several -- commercial and open source -- implementations.
In contrast, his pseudo-assembler is arcane. Whenever I wanted to implement an algorithm following Knuth TACOP, I had to work off his English pseudo-code description rather than the associated pseudo-assembler code.
In short, it's the "mind's 'I'" - we think not as response to external stimuli (only) such as prompts, but we have an inner "I" that asks questions on its own initiative.
There are people like Douglas R. Hofstadter, who believe consciousness is not linked to human hardware (the brain), but that it is an epiphenomenon that emerges as a result of sufficient complexity of the underlying system:
https://en.wikipedia.org/wiki/The_Mind%27s_I
I believe that while underlying high complexity is certainly logically necessary for consciousness, but it is not logically sufficient, and I am undecided (slightly "pro" intuitively) on the question of separability of consciousness from its hardware.
Will a LLM ask an original question on day? I doubt it.
Note that AI models do not have to be conscious to be useful (or to take away millions of jobs)!
Agatha Christie was also a high-volume writer; apparently, this was due to unreasonably demands in her contracts with her publisher, and she hated to be thus pushed.
AI research professor during the day, startup CTO at night, coder, avid reader and book collector.
Wrote my first compiler during high school. Systems I've had a hand in are used by folks ranging from London traders to the Justices of the U.S. Supreme court.