judell

judell, 7 months ago to llm

Here's a test I'll be adding to my repertoire of LLM tests.

"I left Singapore at 9PM on Oct 20, flying east to San Francisco where I will arrive at 9PM on Oct 20. How much daylight will I see looking out the window?"

All of them - ChatGPT4, Claude, Bard - got it spectacularly wrong, confidently asserting I'd see lots of daylight.

When I told them I'm 10 hours into the flight and have seen none so far they were all like "oh, yeah, right, sorry about that."

#llm

judell, 10 months ago to keyboard

"When we use computer keyboards or musical instruments, we need our hands to work unconsciously and automatically. It’s only when injuries happen that we have to surface what’s been unconscious, enter manual mode, and retrain until we can form new unconscious habits. I wish there were an early warning system."

https://blog.jonudell.net/2023/07/23/is-there-an-early-warning-system-for-rsi/

#rsi #keyboard

judell, 1 year ago to random

I'm downloading Descript as I write this. My first intro to text as the interface to audio editing (such a profound and wonderful idea!) was courtesy of Laurian Gridinoc years ago, I've been eagerly awaiting more general availability.

https://cogdogblog.com/2023/05/changing-up-decripting-my-podcast-methods-eh-ai-eh/

See also https://blog.jonudell.net/2022/09/25/curating-the-studs-terkel-archive/ about remixing the Studs Terkel archive using Laurian's https://hyper.audio/. I wonder if Descript uses the same or a different foundation?

#Audio #Transcription #Editing #Remixing

judell, 10 months ago to random

"it may become increasingly difficult to train newer versions of LLMs without access to data that was crawled from the Internet prior to the mass adoption of the technology"

https://arxiv.org/pdf/2305.17493.pdf

Yet another crucial role for the Internet Archive. Its pre-2020 crawls will be used to mitigate model collapse.

judell, 8 months ago to llm

"Some explanations can, will, and should be written by code authors alone, or by those authors in partnership with LLMs. Others can, will, and should be conjured dynamically by code readers who ask LLMs for explanations on the fly."

https://thenewstack.io/how-to-use-llms-for-dynamic-documentation/

#LLM #Documentation #LiterateProgramming

judell, 1 month ago to random

There's much talk of AI-driven unemployment, not so much about AI-powered employment. Reflecting on a week in which I accomplished a lot more than I could have two years ago, I'm feeling conflicted. It's exhilarating to be empowered by tools that make me a more effective thinker and doer. It's also starting to feel exhausting. As with all forms of augmentation, we need to find balance. Just because we can exercise these new superpowers doesn't mean we always should.

judell, 5 months ago to llm

Knowledge graph, ontology, semantic metadata: always nice-to-have but never must-have because the payoff for investing was always speculative.

The argument here: that investment now enables LLMs to write SQL that answers business questions and delivers immedate, repeatable, and measurable benefits.

I never thought RDF was for humans and still don't, but I can see how it could provide the patterns the machines need to be better analytical assistants.

https://thedataexchange.media/knowledge-graphs-llm-sql-rag/

#sql #llm #rdf

judell, 1 year ago to random

The ideal way to write an essay, for me, is to dictate it while on a long hike. You can't beat fresh air and exercise versus sitting at a desk.

I've rarely managed the trick, because it feels awkward to dictate and because it's hard to have all the necessary context in my head.

But when it works, as it did tonight, it feels wonderful!

#Writing #Dictation #Exercise #FreshAir

judell, 11 months ago to random

Yup.

This is the steampipe.io pitch:

"Query like it's 1992
Stop hacking around with scripts; get real work done with the data access standard that's been rocking it for 4 decades."

Of course SQL has evolved in ways many fail to appreciate.

Can you think of a better way to make the point and avoid the eye roll? https://infosec.exchange/@sickmatter/110554514802515393

judell, 2 months ago to random

These two items were adjacent in my feed.

https://fedi.simonwillison.net/@simon/112237545994624170

https://octodon.social/@jalefkowit/112238324590392720

I hugely respect both @simon and @baldur. How to square their opposing views? Two things can be true. Some who use LLMs are experiencing a massive boost. Others are fooling themselves about AGI.

judell, 9 months ago to llm

"Just as ChatGPT can make up facts, it’s apparently willing to lie about ensuring that the code it writes passes the tests you give it. It can also behave like a recalcitrant child who knows, but must constantly be reminded, to follow the rules. But if you hold its feet to the fire, tests can be a great way to focus its attention on the code you’re asking it to write."

#llm #programming #testing

https://thenewstack.io/test-driven-development-with-llms-never-trust-always-verify/

judell, 9 months ago to random

I will not miss the internal combustion engine my next car hopefully will not have.

But I am really going to miss the manual transmissions I've enjoyed for nearly my whole automotive life.

#EVNotYet

judell, 6 months ago to postgres

All the Steampipe plugins are now also:

• standalone Postgres foreign data wrapper (FDW) extensions that translate APIs to foreign tables

• SQLite extensions that transate APIs to virtual tables

• standalone binaries that export data from APIs

It's kind of a big deal! I'm especially curious to know what @simon thinks about all the new SQLite extensions that just appeared in the world.

https://steampipe.io/blog/2023-12-sqlite-extensions

#steampipe #postgres #sqlite

judell, 4 months ago to random

"We are not anti-machine learning," he @cigitalgem] says. "We don't say throw out the baby [with the bath water]. We say, why don't you scrub that baby? Wait a minute, tell us exactly, where did you get the tub?"

https://www.darkreading.com/cyber-risk/researchers-map-ai-threat-landscape-risks

judell, 2 months ago to LLMs

I've been thinking for a long time about tools to help people learn to be better writers. The latest experiment wasn't a resounding success, nor did I really expect that. But it feels promising, and I'm interest to compare notes with fellow travelers. I know wattenberger@bird.makeup is one, who else?

#llms #writing #education

https://thenewstack.io/using-ai-to-improve-bad-business-writing/

judell, 4 months ago to LLMs

"When we can partner with machines that have read the docs, and can look over our shoulders as we try to do the things described in the docs, we’ll turbocharge our ability to dive into unfamiliar software tools and quickly learn how to use them."

https://thenewstack.io/how-to-learn-unfamiliar-software-tools-with-chatgpt/

#llms #LearnByDoing

judell, 1 year ago to fediverse

"It’s a re-calibration to a more human scale I desperately needed. My tweets were getting lost in all the noise for many years, but on Mastodon I’m finding connection again, folks are actually responding, conversations are happening, and there’s a growing sense of community."

#mastodon #community

https://bavatuesdays.com/the-allure-of-mastodon/

judell, 1 month ago to random

Here's one for the @b0rk compendium of git oddities.

Today I learned there is a difference between a lightweight tag:

$ git tag v0.20
$ git cat-file -t v0.20
commit

and an annotated tag:

$ git tag -a v0.21 -m "this is an annotated tag"
$ git cat-file -t v0.21
tag

I have only ever used the former, but it's possible for software to care about the difference between the former (the type of the object is commit) and the latter (the type is tag).

What a tangled web we weave.

judell, 9 months ago to ethelcain

"Bama Rush codifies the many incentives behind marrying power and turns them into a long audition to become a handmaiden to patriarchal privilege."

Not something I'd normally find interesting, but great writing can make any topic come alive.

👏 @tressiemcphd

https://www.nytimes.com/2023/08/22/opinion/bama-rush-tiktok-race.html

#Alabama #Sorority

judell, 6 months ago to random

The word "content" is like fingernails on chalkboard to me, it makes me flinch. I always wish for alternative words like "news article" or "blog post" or "video" or "podcast". Unrealistic? Perhaps, I thought.

Then I asked Claude for 5 sentences that use the word and 5 rewrites that avoid it. Not so hard!

(Forgot to tell it to also avoid "creator".)

#You_Dont_Have_To_Use_The_Word_Content

judell, 7 months ago to random

"I feel a little dirty about those less-than-green travel choices, but I had my reasons this time.

Cost was foremost among them. Amtrak fares on the Northeast Corridor have become crazy expensive"

Am I correct in thinking that this awful incentive affects rail travel in the UK, Europe, and elsewhere as well? https://robpegoraro.com/2023/11/03/cheating-on-amtrak/

judell, 10 months ago to random

The nightmare of self-service has reached the point where I'll try a phone call first in order to avoid another infuriating encounter with a website.

Surprisingly often, I reach a human who is pleasant and helpful. Let's employ more of them. AI can and should assist, but we need to rehumanize customer service. Companies that do will earn my business.

#CustomerService #TechDisservice https://appdot.net/@jgordon/110826817370616225

judell, 5 months ago to books

LibraryThing has been going strong since 2005. I interviewed its founder and lead developer, Tim Spalding, back in 2008. Listening to that interview again today reminded me that everything I loved about LibraryThing remains important and matters even more now.

https://blog.jonudell.net/2023/12/25/critical-mass-in-the-goldilocks-zone/

#books #socialsoftware #criticalmass #librarything

judell, 11 months ago to writing

"The value in writing lies in what we discover while writing."

https://softwarecrisis.dev/letters/bubble-emotions/

#Writing #LLM

/ht @alcinnz

judell, 1 year ago to random

I know it isn't necessary to prefix an LLM prompt with "Please" but I find myself always doing it. Not because I'm trying to anthropomorphize the machines, or at least I hope not, but rather because I don't want to lose the habit of asking for things politely.

#LLM