jurph.bsky.social
Veteran of dial up BBS, mIRC, AOL federated messaging, and Google Reader. When all this is gone I will remain. The most important thing is helping people.
1,122 posts
261 followers
531 following
Regular Contributor
Active Commenter
comment in response to
post
"Sir we air assaulted the regiment's gift shop and brought you back this shirt, we'd be honored if you'd wear it in a photo op"
comment in response to
post
You gotta admit that if an HBCU is going to talk the talk on diversity, admitting a Quebecois student is a helluva way to walk the walk
comment in response to
post
Schumer & Pelosi taking notes for tonight’s fundraising email blast
comment in response to
post
comment in response to
post
comment in response to
post
Oh no RIP my chances of getting one I suppose
I wasn’t tracking the release date!
comment in response to
post
Get you a team that could assemble a TD drive even if the forward pass were outlawed for 10 minutes
comment in response to
post
Covering Fleetwood Mac is an act of hubris. All you can really aspire to is "novel", not "better". You might do something novel enough that for some audiences it seems better, and that's a kind of greatness too, but unless you bring a messy polycule you're not getting over the bar emotionally.
comment in response to
post
Ohhhhh, it was a
joint
column,
no I understand now, yeah, no, yeah.
comment in response to
post
If you can point out that their ceiling is college, and they need to find a major that can leverage their passion - and the show sets them on that path! - what fun, what delight.
comment in response to
post
You get the parents and teams to buy in because there will be professional scouts and pro athletes there, WITH CAMERAS. You show highlights of the kids... then highlights of that athlete (even split-screen slo-mo at the same time?). Then take the parents aside and say "his ceiling is a 1-AA school".
comment in response to
post
He's literally the Commander in Chief. Congress has no power to stop him from issuing lawful orders.
I hate his domestic politics, I hate his ignorance, I hate his hateful policies, but this is the issue on which Trump's opponents have the very weakest hand.
Fight literally anything else.
comment in response to
post
modern problem
require
modern solution
comment in response to
post
Literally just playing D&D with the engineers. “So you’re Dave and you wear a spacesuit… and I’m HAL and you’re going to fix the spaceship? Okay uhhh that sounds grim but I’m in. Better check the exterior cooling systems. Pod bay is right there.”
comment in response to
post
One of my tutors from back in the 90’s told us: as soon as your employer doesn’t make payroll- get out!
comment in response to
post
comment in response to
post
“MANEUVER!”
comment in response to
post
This cloud looks like Yoda to me.
comment in response to
post
"queso wormhole" was the name of my TMBG mariachi cover band in college
comment in response to
post
"Have a plan, and stick to it."
comment in response to
post
comment in response to
post
I can’t believe you’d mute such an important issue! You must be pro-FRAUD or a false flag shill for the Carceral State, you heartless yogurt! It’s our sacred right to launder money and/or report people for laundering money to Schenectady orphanages full of deepfake cardboard waifs - are you mad?!
comment in response to
post
I've even got teammates who are privileged enough to be able to train from-scratch models on non-infringing datasets! I know this is not something everyone can do (and few should bother) but there is a path to lightweight non-infringing models that protect user privacy.
comment in response to
post
Which is sort of why I was cautioning against the earlier absolutist rhetoric - you and I share a common goal (greatly reduce energy usage, improve privacy) but I am also a deliberate and enthusiastic user of GenAI. We ought to be partners in this, but my work is demonized along with the worst uses.
comment in response to
post
Oh, for sure - it's by far the most prominent use case. And treating GPTs like search engines is a whole other problem.
I am working in the area of tool use and agentic systems, though, where small local models are favored, and reducing energy consumption and hallucinations is a major goal.
comment in response to
post
Broadly, though, yes - anything you put into a query on the website or the app probably will get used as part of next-gen training data. ChatGPT and Claude are only available in that mode, and they're what people are most familiar with. But there's a lot of work with lightweight local models.
comment in response to
post
It's not strictly that simple - if you're querying a remote model (Software as a Service, SaaS, etc.) then they log the queries, and feed them into a training set for the next-gen model. But you can also download models like LLAMA, Qwen, etc. and run them locally, and nothing is logged.
comment in response to
post
I understand the online SaaS models use session data to train on, which is part of why I prefer locally hosted models that don’t leak my queries.
I think I’m missing how it makes future thefts easier; my understanding was that the scraping of training corpora is constant and ongoing.