danielbln

@danielbln@lemmy.world

This profile is from a federated server and may be incomplete. Browse more on the original instance.

danielbln,

I’ve implemented a few of these and that’s about the most lazy implementation possible. That system prompt must be 4 words and a crayon drawing. No jailbreak protection, no conversation alignment, no blocking of conversation atypical requests? Amateur hour, but I bet someone got paid.

danielbln, (edited )

Depends on the model/provider. If you’re running this in Azure you can use their content filtering which includes jailbreak and prompt exfiltration protection. Otherwise you can strap some heuristics in front or utilize a smaller specialized model that looks at the incoming prompts.

With stronger models like GPT4 that will adhere to every instruction of the system prompt you can harden it pretty well with instructions alone, GPT3.5 not so much.

danielbln,

Eh, that’s not quite true. There is a general alignment tax, meaning aligning the LLM during RLHF lobotomizes it some, but we’re talking about usecase specific bots, e.g. for customer support for specific properties/brands/websites. In those cases, locking them down to specific conversations and topics still gives them a lot of leeway, and their understanding of what the user wants and the ways it can respond are still very good.

Platformer is leaving Substack for Ghost, an Open-Sourced publishing platform (www.platformer.news)

“After much consideration, we have decided to move Platformer off of Substack. Over the next few days, the publication will migrate to a new website powered by the nonprofit, open-source publishing platform Ghost. If you already subscribe to Platformer and wish to continue receiving it, you don’t need to do anything: your...

danielbln,

Better host it yourself, because the managed version of Ghost has the lovely dark pattern that you cannot cancel your plan without them immediately deleting everything, no matter if your plan is still paid for for another few months or not. Really left a bad taste on my mouth.

danielbln,

Only black or green tea, herbal tea (sorry, infusion) not so much.

danielbln,

I’m gonna saythat those aren’t tiles but some shitty tile print linoleum.

danielbln,

Have you tried not eating crappy Falafel?

danielbln,

And that’s just smart design, so you don’t get fried if the breaker or RCD is physically blocked or is wilfully held in place, like with hillbilly Bob over here.

danielbln,

Microsoft has shown with Phi-2 (microsoft.com/…/phi-2-the-surprising-power-of-sma…) that synthetic data generation can be a great source for training data.

danielbln,

Good luck on foot, brah. I’ll be taking the jeep and hose-sucking fuel where needed. Provides shelter, warmth, storage.

danielbln,

They might also not enjoy going down with their claws tied.

I feel like the Steam Deck is the best proof of Gabe Newell's quote that "piracy is a service issue."

They could have easily crammed the Steam Deck full of stuff to make it hard to use for piracy - locking down everything, making it usable only to play games you legitimately own, force you to go through who knows what hoops in order to play games on it. That’s what Nintendo or Apple or most other companies do....

danielbln,

Let’s just hope for Gabe to live a long life still. Valve is a private company and not nearly as much in danger for enshittification as a public company would be.

danielbln,

But no, not like that. Clearly they need to find a way of materializing money from thin air.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • localhost
  • All magazines
  • Loading…
    Loading the web debug toolbar…
    Attempt #