Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

LaMDA: Google's New Conversation Technology

blog.google

Post reply on HN
Print view View original

81–90 of 157 posts

Re: LaMDA: Google's New Conversation Technology

#81

Post by TulliusCicero »

raphlinus wrote: post #52

Disclosure: work at Google, opinions my own. I've had a chance to play with this and it's eerie how good it is. Most impressive to me, it was able to defend itself persuasively when confronted with evidence of how its previous answers were inconsistent (in the version I played with, it tends to be very coherent but frequently wrong).

Also a Googler, and same.

I tried to get it to say controversial/inflammatory/bigoted things with leading questions, or questions where the framing itself was problematic, and it was quite adept at deflecting/dodging/resisting. I was very surprised.

Re: LaMDA: Google's New Conversation Technology

#82

Post by ctoth »

joe_the_user wrote: post #62
raphlinus wrote: post #52

Disclosure: work at Google, opinions my own. I've had a chance to play with this and it's eerie how good it is. Most impressive to me, it was able to defend itself persuasively when confronted with evidence of how its previous answers were inconsistent (in the version I played with, it tends to be very coherent but frequently wrong).

Does it have to defend itself a lot? I've noticed contradictory statements are kind of standard for GPT-3. I can well believe you could come up with a system that's better at defending itself against the charge of contradiction than it is at avoiding contradictions in the first place.

I've noticed contradictory statements are kind of standard for humans. I can well believe you could come up with a system that's better at defending itself against the charge of contradiction than it is at avoiding contradictions in the first place.

Re: LaMDA: Google's New Conversation Technology

#83

Post by TulliusCicero »

joe_the_user wrote: post #62
raphlinus wrote: post #52

Disclosure: work at Google, opinions my own. I've had a chance to play with this and it's eerie how good it is. Most impressive to me, it was able to defend itself persuasively when confronted with evidence of how its previous answers were inconsistent (in the version I played with, it tends to be very coherent but frequently wrong).

Does it have to defend itself a lot? I've noticed contradictory statements are kind of standard for GPT-3. I can well believe you could come up with a system that's better at defending itself against the charge of contradiction than it is at avoiding contradictions in the first place.

> Does it have to defend itself a lot?

I think it's a common weakness of chatbots that you can get them to contradict themselves through presenting new evidence or assertions, or even just through how you frame a question. I found LaMDA to be much more resistant to this than I anticipated, when I tried it out (I'm a Googler).

It wasn't completely immune -- I was eventually able to get it to say that, indeed, Protoss is OP -- but it took a long time.

Re: LaMDA: Google's New Conversation Technology

#84

Post by ksenzee »

anonagoog wrote: post #26
epr wrote: post #19

We really need to stop calling things "lambda". You'd think Google would grasp the concept of unique names on the internet, but I suppose they can always push their own stuff to the top.

This anodyne name is likely because there was a small internal shitstorm over calling it Meena because some folks were tired of chat bots being given women’s names

As a woman I appreciate it. I get so tired of things we order around being named after women.

Re: LaMDA: Google's New Conversation Technology

#85

Post by shawndumas »

zamadatix wrote: post #80
waynesonfire wrote: post #69

Earlier quoted context omitted.

Sure, you should be cognizant of what you say online otherwise you'll get smacked with the cancel culture hammer. This is regardless of where you work.

The net of what you should be cognizant of when providing insider information on a product from your company or commenting on a topic it is heavily involved in is wider than the single issue you eagerly dredge up. E.g. the last thing you want from your comment about using early internally are news articles about "What Google plans on doing with " when you were just trying to give your personal opinion on where the te…

To echo this comment, zamadatix is exactly right; I need someone quoting me about a Google project as an official representative about as much as I need another hole in my head.

As to the greater point, I want to emphasize that as a human being I want to be careful with how I communicate with others because I care about them not feeling judged, dismissed, othered, or otherwise insulted. The charge that the rules keep changing is overblown; it’s always about respecting people even if you don’t understand or enjoy them.

I know that some might still equate that with “cancel culture” but I can’t help that (and tbh that’s a discussion that we should have over beers and not on the internet).

Re: LaMDA: Google's New Conversation Technology

#86

Post by ocdtrekkie »

ksenzee wrote: post #84
anonagoog wrote: post #26

Earlier quoted context omitted.

This anodyne name is likely because there was a small internal shitstorm over calling it Meena because some folks were tired of chat bots being given women’s names

As a woman I appreciate it. I get so tired of things we order around being named after women.

IMHO, we probably should try to get to the point that bots/assistants are less "a thing we order around" and "a thing we communicate with", if partially because of the pretty bad vibes of "creating slaves" or being comfortable ordering such programs around, regardless of the degree we are confident we haven't accomplished computer sentience. It's just a bad pattern to be in. I recall a lot of work going into making Cortana refuse to tolerate abusive language towards it.

Maybe Google has it right with not anthropomorphizing Google Assistant and giving it a human name, but in that case LaMDA is a move in the wrong direction it seems: The conversations demonstrated emulated emotion in a way that generally speaking, most bots do not.

Is there any work on making difficult-to-gender voices for bots that still sound human? Maybe if we were to combine names that aren't traditionally given to humans with a voice that is hard to gender? Should we intentionally make bots/assistants sound more robotic, so they do not try to "sound human"?

I think it's kinda an interesting question to figure out what the best practice should be moving forwards.

Re: LaMDA: Google's New Conversation Technology

#87

Post by lallysingh »

raphlinus wrote: post #52

Disclosure: work at Google, opinions my own. I've had a chance to play with this and it's eerie how good it is. Most impressive to me, it was able to defend itself persuasively when confronted with evidence of how its previous answers were inconsistent (in the version I played with, it tends to be very coherent but frequently wrong).

So, tried Turing testing it vs some less-socially-adept googlers?

Discloser: xoogler here

Re: LaMDA: Google's New Conversation Technology

#88

Post by tamrix »

epr wrote: post #19

We really need to stop calling things "lambda". You'd think Google would grasp the concept of unique names on the internet, but I suppose they can always push their own stuff to the top.

Marketing would have made this name up.

The main market for Google tech are brogrammers who would buy anything with buzzwords in it.

Re: LaMDA: Google's New Conversation Technology

#89

Post by lupire »

Traster wrote: post #73
shawndumas wrote: post #67

Earlier quoted context omitted.

Disclosure: I work at Google too and my opinions are also my own. We are told internally that when we discuss Google related matters online that we should be cognizant of the what/where/how/why of our communication. Disclosing is—for me—both a reminder for myself and an attempt at a hedge against accidentally representing Google.

FWIW Intel engineers are meant to use the hashtag #iamintel to indicate their role, see it a lot on LinkedIn

Ew. Working for Intel isn't being Intel.
Post reply on HN