ChatGPT Flirts Now? | Ep 83
Вставка
- Опубліковано 16 чер 2024
- This week, OpenAI unveiled GPT-4o, its newest A.I. model. It has an uncannily emotive voice that everybody is talking about. Then, we break down the biggest announcements from Google IO, including the launch of A.I. overviews, a major change to search that threatens the way the entire web functions. And finally, Kevin and Casey discuss the weirdest headlines from the week in another round of HatGPT.
Chapters
00:00:00 - Coming up
00:01:11 - This Week on Hard Fork
00:00:59 - Kevin at War with a car wash
00:03:43 - ChatGPT’s new Voice / Updates
00:29:36 - Google I/O made us sad!
00:54:06 - HatGPT
00:55:11 - FTX Recovered lost money
00:57:21 - Bumble CEO: AI Dating Concierge
01:00:03 - Sonos app sucks
01:02:17 - NYC/Dublin Portal shut down
01:04:02 - Super Mario 28 year old door finally opened!
01:05:27 - Man fools Waymo cars with stop sign t-shirt
01:07:49 - Start telling your friends about us!
Additional Reading:
A.I.’s ‘Her’ Era Has Arrived
www.nytimes.com/2024/05/14/te...
ChatGPT Gets an Emotional Upgrade
www.platformer.news/openai-gp...
Google’s Broken Link to the Web
www.platformer.news/google-io...
Can Google Give A.I. Answers Without Breaking the Web?
www.nytimes.com/2024/05/14/te...
Hard Fork is a weekly look into the future that's already here. Hosts Kevin Roose and Casey Newton explore stories from the bleeding edge of tech.
Casey’s publication, Platformer: www.platformer.news/
FOLLOW US ON SOCIAL
TikTok:www.tiktok.com/@hardfork?is_f...
Insta:
Kevin Roose- kevinroose...
Casey Newton - crumbler?h...
Twitter (X):
Kevin - x.com/kevinroose?s=20
Casey - x.com/CaseyNewton?s=20
Threads:
Kevin - www.threads.net/@kevinroose
Casey - www.threads.net/@crumbler
Subscribe to the audio only podcast:
Apple - podcasts.apple.com/us/podcast...
Spotify - open.spotify.com/show/44fllCS...
Amazon - music.amazon.com/podcasts/7c7...
Google - podcasts.google.com/feed/aHR0...
The New York Times - www.nytimes.com/column/hard-fork
Credits
“Hard Fork” is hosted by Kevin Roose and Casey Newton. Produced by Rachel Cohn and Whitney Jones . Edited by Jen Poyant. Engineering by Alyssa Moxley and original music by Dan Powell, Elisheba Ittoop and Marion Lozano. Our audience is Nell Gallogly. Video production by Ryan Manning and Dylan Bergeson. Motion graphics by Phil Robibero Thumbnails by Julia Moburg, Elizabeth Bristow, and Harshal Duddalwar
Special thanks to Paula Szuchman, Pui-Wing Tam, Nell Gallogly, Kate LoPresti and Jeffrey Miranda. - Наука та технологія
Hands down the best podcast for this type of stuff. Easy to digest, down to earth, funny, relatable, informative, asking the right questions… insanely underrated. Ditch the Times logo and create a Hard Fork logo and really push this shit
Casey's point about how they're working to make it human via little design choices, manipulation in tone or word choice, adding phrases. I work in app dev, marketing tech - and this honestly has been a mainstay of user experience design for years at this point. Chat bots, voice agents, but even workflows and forms when overdesigned leverage elements of this. OpenAI is just going whole hog.
How are there only 9 comments on this video? This is the best podcast I've listened to since the original Radiolab. Thanks.
This was one of the best episodes of the hard fork so far. I mean kudos to you guys, I’m incredibly moved with the content conversation smartly popped jokes.. I mean that was terrific!! Thank you thank you!!!!
Hey Kevin! You guys should cover the world of robotic. There’s so many interesting progress going on in that space including AI built in many of these high end robots.
Using it now with my AirPods - you can literally just have it so that you can talk to it throughout the day. It’s actually insane
I know that open AI was aiming for Scarlett vibes but the voice is kind of a dead ringer for Lola Rabbit. If that doesn’t ring a bell, UA-cam it.
Agreed. The flirting and giggling does not make it a better bot. Still hallucinates. Still does ridiculous refusals. It needs to be useful first and foremost.
Organizations with vested interest in AI typically put research first before product for a long time, until OpenAI decided it would be really neat to have that really dumb baby foundation of what first resembles Scifi AI as a website like askJeeves. Because of course Silicon valley will try to make a buck ahead of all else. Now that's set the race running and has Google absolutely embarrassing itself with failed product launches in the AI space that's no doubt retracting members away from meaningful research
The mention of using it while driving gives me the idea it can be used to make sure a driver doesn’t fall asleep while driving by having something to talk with and it can watch your face to know if the driver starts dozing off
It makes sense on the persona side of things, there is adoption gaps (particularly in Enterprise environments) where often AI interaction would only occur if prior trust was established.
Building trust isn't just "Prompt goes in, answer comes out", its about how that answer is delivered, detailed and diverse in its representation (UX & UI), which the voice aspects natively support. OpenAI has made a lot of moves this week, I would say its the week of "removing roadblocks", They open ChatGPT for free (to get a stable influx of quality training data crowdsourced from millions of new users) which is exactly what you need for AGI. They add Voice and empathic modulation within the same model (which is what you need to properly decode emotional inputs, rather then just TTS where 80% of the intent is lost because you don't have the emotions of how the content was delivered) and opens up the new UX modes for using it (Vision and Voice). Even from the API side: higher limits, lower cost, lower latency, relaxed rules (thanks to model spec), batch processing, scheduled processing... priming the way for autonomous agents which is the next frontier. If you look at "AGI" having 5 key themes (Text, Vision, Voice, Reasoning, Autonomy) then everything OpenAI is doing is in line with their stated mission.
One thing that I think was overlooked in the Bumble/bot dates story is that dating is a learning and growth process - even if a date or relationship doesn’t last long, ideally the participants learn something about themselves, social skills, or even just another perspective of the world. By reducing yourself to a bot that will find a match for you suggests that you’re a static person and takes away that opportunity to develop and change.
$1000 for a car key? Dear lord.
Great episode! ❤
Funny and honest. Great pod.
Very much enjoy your very informative podcast 💛, thank you! 👍
Thanks for calling out Sonos in this episode. I thought I was crazy, it used to be so great but each update it seems to work less and less. And the volume sliders…wth? Hoping for a deeper dive on this baffling company someday.
car keys, even fobs, usually cost less than $50. The programming can be done for free.
The trick is finding stores that won't gouge the customer. I worked at a Batteries Plus, and they started doing car when i left. I could get a fob fully programmed and cut for under $60 for 90% of toyotas and hondas.
Sketchy af tho...i bricked a couple's 2014 accord one time and the actual locksmith had to come fix it. Turns out our key cutting machine ran on hopes and dreams (i knew this) and the cuts were made slightly wrong which triggered the anti-theft system.
I think part of the problem with the idea that the internet is supported by people using Google to get to websites to click on ads is... when was the last time anyone clicked on an ad? I've never, as far as I can remember, have ever clicked on an advertisement on a website that I got to from a Google search. I don't click on ads, I barely even notice them and I don't use an ad blocker. So, to suggest that the internet is going to die because people who are googling for information have stopped having to click on the pages... I don't fully understand it, because I don't really believe there are people who click on ads anyway. It makes more sense to me that ad networks are supported entirely by fraud of some kind, where no organic traffic is at all involved anymore.
I love this podcast and look forward to it every week, but I’d like to hear more about other AI advances that are less “consumer-facing”, how generative AI is being used in science where whether or not it talks is not important.
I would be very interested in that too
you guys are overlooking that this is the first model that tokenizes all data types in a single model.
Did Casey do some ADR to say "a week" at 23:16, or did you use voice synthesis?
that was a fact checked pickup recorded later! so yes, adr.
[for total transparency, he said "a month," originally]
@@hardforkLove a good fact check
A hat tip to the video editor of this podcast! Great work 👏
Has anyone else noticed a drop in audio quality for this week's _podcast_ episode?
I actually came here to see if their recording location or setup had changed (it doesn't seem to have). Interestingly, the UA-cam episode's audio quality sounds great as usual. It's just the _regular_ podcast feed version of this episode (Apple Podcasts, Pocket Casts, etc.) sounds thinner and more compressed.
AGI isn’t sentience. Sentience is the next step beyond
"I wasn't crying at Google i/o" ^^
Fast forward a few weeks, and Casey is sad because he only got to talk for 40 minutes with Chat GPT's new voice mode before it was taken away. I believe you could see firsthand why people would want such a personal assistant. There's nothing like the firsthand experience, right 😂😂😂
AI vertigo ... that's dead-on. Was trying to describe this to someone earlier - feeling out of sorts, like I thought I had things sussed but suddenly felt unmoored after the 4o demo
The ability of the AI to describe a scene or person is actually something that'd be extremely useful for vision impaired people. Having said that, this entire thing is still creepy af. Also, I wish someone would ask at every opportunity, how much energy is this thing using per query?
Love the episode. Don’t think it is media navel gazing. The A.I. clips make me yawn. I’m here for the car key and live nation stories.
Could you fall in love with A.I?
Nope. I think very few people could. I could use it to talk to as an assistant or something like that and it’s nice that it has a pleasant voice, but I could never sit and have a personal conversation with it. What would be the point, you know it doesn’t really care or understand.
A lot of people already have, myself included
What happens tp Hardfork when Google uses a video summariser on this video or uses a summariser on the Spotify' podcast to include the salient points into its AI summaries. I for one will miss the jokey banter - FT Key
I think you might be wrong about your comments that they have programmed in things like pauses, halting words and intonation to make the voice more human. That would be hard coded programming and not generative AI. I suppose you could take the AI output and modify it with code to edit those things in but that seems very unlikely. I think the AI is naturally doing those things because it is trained on real people speaking.
Yay! First view and comment on my favorite podcast! First time!🎉
Regarding publishers, I don't think you can blame Google. Unfortunately, Google is reacting to Bing/ChatGPT, Tiktok, Facebook, etc.
These AI Search will cost google more. If it is not because of competition google would not want to change. The reason google seems to be behind OpenAI because google kept these AI in research and not change their model.
Google is not gaslighting media. Investors don't want to change..Google is feeling pressure from companies like perplexity
Something to consider also is that the more OpenAI focuses on the search being multi-modal and not requiring people to look at a screen the harder it will be for Google to compete with that. Google needs people to look at and click the paid ads. So the attractive flirty voice is a clear disruption for Google.
It will be illegal to wear a STOP sign on T shirts, soon. Either that, or Tesla FSD will have to train itself on that edge case.
Always has
I have not gone to a website for a news story in years. The amount of ads and popups is unbearable. This will be positive for the consumer.
Try the New York Times. Almost no ads.
@@TheCassady34 no
I find this unsettling. I still need the assistance from an artificial intelligence, to be characterized by a distinctly robotic and non-human nature.🙉
Great podcast. Something that was claimed that I'm not sure is true is that OpenAI DESIGNED its demo chatbot voice to be flirty, manipulative, friendly, complimentary, whatever. That's plausible, but I wonder if that's just a downstream effect of training multimodal models natively on voice. It seems just as plausible that it's merely a reflection of training data. If they used something like RLHF for 4o, maybe the human feedback people just tended to prefer responses in that cheery emotional mode.
Anyway, I'll be pacing back and forth in my room until they release an AI that can keep me constantly hypnotized in mindless bliss.
OpenAI's voice is so cringe. This is NOT what I want in an AI assistant. Bridget at CNET had a great rant about it 😅
It has other voice options I think.