It's really how it works.
Winner of the 'understatement of the week' award (and it's only Monday).
Also top contender in the 'technically correct' category.
Yes! As soon as I saw gdb I was like "that can't be Greg", but sure enough, that's him.
Seems like these would be similar.
This really isn't surprising. Look at Google Home and Alexa. When they first came out, if you weren't a white male from the west coast... the accuracy of translating commands dropped DRAMATICALLY. Because it was programmed, designed and tested by majority of white tech bros in SF/Seattle. They've gotten a lot better over the last 5+ years. But I think you'll see OpenAI take this route.
But that's ok. They have to start somewhere. Once they get the model working really well for one language they can expand into similar ones with relatively little work. The more different the language, the more hard work and "local input" (ex: natives of the language) will be required for adaptation. But the basic text translations are still already way better than they used to be.
It's a shame, because that means I can use stuff that I can't recommend to people around me
Multilingual UX is an interesting painpoint, I had to change the language of my account to English so I could use some early Bard version, even though It was perfectly able to understand and answer in Spanish
(accents sure)
(Siri's original speech recognition was a combination of Swiss-Germans and people from Boston.)
And it certainly wouldn't be tested by them either way. Companies know how to hire QA contractors.
Even beside that though -- it's patently ridiculous to suggest that these devices would perform worse with an Asian man who speaks fluent English and was born in California. Or a white woman from the Bay Area. Or a white man from Massachusetts.
You kind of have a point about tech being the product of the culture in which it was produced, but the needless exaggerated references to gender and race undermine it.
But for conversation, you can’t speak Mandarin without using proper tones because you simply won’t be understood.
You can be perfectly understood if you don't speak broadcast Chinese. There are plenty of heavy accents to deal with anyways. Like Beijing 儿化 or the inability of southerners to pronounce sh very differently from s.
What's strange to me is that, as far as I know, "plátano" is only commonly used in Spain, but the accent of the AI voice didn't sound like it's from Spain. It sounds more like an American who speaks Spanish as a second language, and those folks typically speak some Mexican dialect of Spanish.
The wiktionary page for "plátano" has a map illustrating how various Spanish-speaking countries refer to the banana.
https://en.wiktionary.org/wiki/pl%C3%A1tano#/media/File:Porp...
My principal association with plátano is plaintain, personally, but I am not a Spanish speaker.
Plátano is commonly used for banana in Mexico, just bought some at a Soriana this weekend.
And note that understanding broken language is a skill you have to train. If you aren't used to it then it is impossible to understand what they say. You might not have been in that situation if you are an English speaker since you are so used to broken English, but it happens a lot for others.
The Italian sounded good to me.
The first (and imo the main) hurdle is not reproduction, but just learning to hear the correct sounds. If you don't speak Hindi and are a native English speaker, this [1] is a good example. You can only work on nailing those consonants when they become as distinct to your ear as cUp and cAp are in English.
We can get by by falling back to context (it's unlikely someone would ask for a "shit of paper"!), but it's impossible to confidently reproduce the sounds unless they are already completely distinct in our heads/ears.
That's because we think we hear things as they are, but it's an illusion. Cup/cap distinction is as subtle to an Eastern European as Hindi consonants or Mandarin tones are to English speakers, because the set of meaningful sounds distinctions differs between languages. Relearning the phonetic system requires dedicated work (minimal pairs is one option) and learning enough phonetics to have the vocabulary to discuss sounds as they are. It's not enough to just give feedback.
interestingly, i think this isn't always true -- i was able to coach my native-spanish-speaking wife to correctly pronounce "v" vs "b" (both are just "b" in spanish, or at least her dialect) before she could hear the difference; later on she was developed the ability to hear it.
Even styles of thought might be different in other languages, so I don't say that lightly... (stay strong, Sapir-Wharf, stay strong ;)
I helped her access the video from the presentation, and it brought her to tears. Now, she can play guitar, and the AI and her can write songs and sing them together.
This is a big day in the lives of a lot of people whom aren't normally part of the conversation. As of today, they are.
Eventually it would be better for these models to run locally from a security point if view, but this is a great first step.
And here are some ideas I put together around 2010 on how to deal with the socio-economic fallout from AI and other advanced technology: https://pdfernhout.net/beyond-a-jobless-recovery-knol.html "This article explores the issue of a "Jobless Recovery" mainly from a heterodox economic perspective. It emphasizes the implications of ideas by Marshall Brain and others that improvements in robotics, automation, design, and voluntary social networks are fundamentally changing the structure of the economic landscape. It outlines towards the end four major alternatives to mainstream economic practice (a basic income, a gift economy, stronger local subsistence economies, and resource-based planning). These alternatives could be used in combination to address what, even as far back as 1964, has been described as a breaking "income-through-jobs link". This link between jobs and income is breaking because of the declining value of most paid human labor relative to capital investments in automation and better design. Or, as is now the case, the value of paid human labor like at some newspapers or universities is also declining relative to the output of voluntary social networks such as for digital content production (like represented by this document). It is suggested that we will need to fundamentally reevaluate our economic theories and practices to adjust to these new realities emerging from exponential trends in technology and society."
And a related YouTube video: "The Richest Man in the World: A parable about structural unemployment and a basic income" https://www.youtube.com/watch?v=p14bAe6AzhA "A parable about robotics, abundance, technological change, unemployment, happiness, and a basic income."
My sig is about the deeper issue here though: "The biggest challenge of the 21st century is the irony of technologies of abundance in the hands of those still thinking in terms of scarcity."
Technology has leapfrogged nature and our consumption patterns have not caught up to modern abundance. Scott Galloway recently mentioned this in his OMR speech and speculated that GLP1 drugs (which actually help addiction) will assist in bringing our biological impulses more inline with current reality.
In short, humans are adapted for a world with a scarcity of salt, refined carbs like sugar, fat, information, sudden motion, and more. But the world most humans live in now has an abundance of those things -- and our previously-adaptive evolved inclinations to stock up on salt/sugar/fat (especially when stressed) or to pay attention to the unusual (a cause of stress) are now working against our physical and mental health in this new environment. Thanks for the reference to a potential anti-addiction substance. Definitely something that deserves more research.
My sig -- informed by the writings of people like Mumford, Einstein, Fuller, Hogan, Le Guinn, Banks, Adams, Pet, and many others -- is making the leap to how that evolutionary-mismatch theme applies to our use of all sorts of technology.
Here is a deeper exploration of that in relation to militarism (and also commercial competition to some extent): https://pdfernhout.net/recognizing-irony-is-a-key-to-transce... "There is a fundamental mismatch between 21st century reality and 20th century security thinking. Those "security" agencies are using those tools of abundance, cooperation, and sharing mainly from a mindset of scarcity, competition, and secrecy. Given the power of 21st century technology as an amplifier (including as weapons of mass destruction), a scarcity-based approach to using such technology ultimately is just making us all insecure. Such powerful technologies of abundance, designed, organized, and used from a mindset of scarcity could well ironically doom us all whether through military robots, nukes, plagues, propaganda, or whatever else... Or alternatively, as Bucky Fuller and others have suggested, we could use such technologies to build a world that is abundant and secure for all. ... The big problem is that all these new war machines and the surrounding infrastructure are created with the tools of abundance. The irony is that these tools of abundance are being wielded by people still obsessed with fighting over scarcity. So, the scarcity-based political mindset driving the military uses the technologies of abundance to create artificial scarcity. That is a tremendously deep irony that remains so far unappreciated by the mainstream."
Conversely, reflecting on this more just now, are we are perhaps evolutionarily adapted to take for granted some things like social connections, being in natural green spaces, getting sunlight, getting enough sleep, or getting physical exercise? These are all things that are in increasingly short supply in the modern world for many people -- but which there may never have been much evolutionary pressure previously to seek out, since they were previously always available.
For example, in the past humans were pretty much always in face-to-face interactions with others of their tribe, so there was no big need to seek that out especially if it meant ignoring the next then-rare new shiny thing. Johann Hari and others write about this loss of regular human face-to-face connection as a major cause of depression.
Stephen Ilardi expands on that in his work, which brings together many of these themes and tries to help people address them to move to better health.
From: https://tlc.ku.edu/ "We were never designed for the sedentary, indoor, sleep-deprived, socially-isolated, fast-food-laden, frenetic pace of modern life. (Stephen Ilardi, PhD)"
GPT-4o, by apparently providing "her" movie-like engaging interactions with an AI avatar that seeks to please the user (while possibly exploiting them) is yet another example of our evolutionary tendencies potentially being used to our detriment. And when our social lives are filled-to-overflowing with "junk" social relationships with AIs, will most people have the inclinations to seek out other real humans if it involves doing perhaps increasingly-uncomfortable-from-disuse actions (like leaving the home or putting down the smartphone)? Not quite the same, but consider: https://en.wikipedia.org/wiki/Hikikomori
Related points by others:
"AI and Trust" https://www.schneier.com/blog/archives/2023/12/ai-and-trust.... "In this talk, I am going to make several arguments. One, that there are two different kinds of trust—interpersonal trust and social trust—and that we regularly confuse them. Two, that the confusion will increase with artificial intelligence. We will make a fundamental category error. We will think of AIs as friends when they’re really just services. Three, that the corporations controlling AI systems will take advantage of our confusion to take advantage of us. They will not be trustworthy. And four, that it is the role of government to create trust in society. And therefore, it is their role to create an environment for trustworthy AI. And that means regulation. Not regulating AI, but regulating the organizations that control and use AI."
"The Expanding Dark Forest and Generative AI - Maggie Appleton" https://youtu.be/VXkDaDDJjoA?t=2098 (in the section on the lack of human relationship potential when interacting with generated content)
[1] https://www.amazon.com/Het-geheim-verdwenen-mysterie-Dutch/d...
I don't need to imagine that, I've had it for about 8 years. It's OK.
> help you grocery shop without an assistant
Isn't this something you learn as a child? Is that a thing we need automated?
Beautiful articulation.
This is an enormous win for humanity.
Yeah, real win for humanity, and not the psycho AI sycophants
Ah yes, also known as being co-founder :)
Even if that were true¹, it doesn’t mean everyone would know their HN user name.
¹ Greg may be well known within a select group of people but that’s way smaller than even users of ChatGPT.
Greg Brockman is famous for good reasons but constant "oh wow it's Greg Brockman" are noisy.
As a language learner, this would be tremendously useful.
You can use any open source model wirthout any promot whatsoever
(GPT) Models toggle between a state of existence when queried and ceasing to exist when not.
That’s fundamentally not how GPT models work, but you can easily build a framework around them that calls them in a loop; you’d need a special system prompt to get anything “thought like” that way, and if you want it to be anything other than stream-of-simulated-consciousness with no relevance to anything, and a non-empty “user” prompt each round, which could be as simple as time, a status update on something in the world, etc.
I suppose it would cost even more electricity to have ChatGPT musing alone though, burning through its nvidia cards...
They are designed for query and reponse. They don't do anything unless you give them input. Also there's not much research on the best architecture for running continuous though loops in the background and how to mix them into the conversational "context". Current LLMs only emulate single thought synthesis based on long-term memory recall (and some goes off to query the Internet).
> I think when I'm alone without query from another human.
You are actually constantly queried, but it's stimulation from your senses. There are also neurons in your brain which fires regularly, like a clock that ticks every second.
Do you want to make a system that thinks without input? Then you need to add hidden stimuli via a non-deterministic random number generator, preferably a quantum based RNG (or it won't be possible to claim the resulting system has free-will). Even a single photon hitting your retina can affect your thoughts and there are no doubt other quantum effects that trips neurons in your brain above the firing threshold.
I think you need at least three of four levels of loops interacting, with varying strength between them. First level would be the interface to the world, the input and output level (video, audio, text). Data from here are high priority and is capable of interrupting lower levels.
The second level would be short term memory and context switching. Conversations needs to be classified, and stored in a database, and you need an API to retrieve old contexts (conversations). You also possibly need context compression (summarization of conversations in case you're about to hit a context window limit).
The third level would be the actual "thinking", a loop that constantly talks to itself to accomplish a goal using the data from all the other levels but mostly driven by the short term memory. Possibly you could go super-human here and spawn multiple worker processes in parallel. You need to classify the memories by asking; do I need more information? where do I find this information? Do I need an algorithm to accomplish a task? What is the completion criteria. Everything here is powered by an algorithm. You would take your data and produce a list of steps that you have to follow to resolves to a conclusion.
Everything you do as a human to resolve a thought can be expressed as a list or tree of steps.
If you've had a conversation with someone and you keep thinking about it afterwards, what has happened is basically that you have spawned a "worker process" that tries to come to a conclusion that satisfies some criteria. Perhaps there was ambiguity in the conversation that you are trying to resolve, or the conversation gave you some chemical stimulation.
The last level would be subconscious noise driven by the RNG, this would filter up with low priority. In the absence of other external stimuli with higher priority, or currently running thought processes, this would drive the spontaneous self-thinking portion (and dreams) when external stimuli is lacking.
Implement this and you will have something more akin to true AGI (whatever that is) on a very basic level.
I think you can get models to "think" if you give them a goal in the system prompt, a memory of previous thoughts, and keep invoking them with cron
You could say, in a sense, that without a human mind to collapse the wave function, the superposition of data in a neural net's weights can never have any meaning.
Even when we build connections between these statistical systems to interact with each other in a way similar to contemplation, they still require a human-created nucleation point on which to root the generation of their ultimate chain of outputs.
I feel like the fact that these models contain so much data has gripped our hardwired obsession for novelty and clouds our perception of their actual capacity to do de novo creation, which I think will be shown to be nil.
An understanding of how LLMs function should probably make this intuitively clear. Even with infinite context and infinite ability to weigh conceptual relations, they would still sit lifeless for all time without some, any, initial input against which they can run their statistics.
Sorry to hijack, but how the hell can I solve this? I have the EXACT SAME error on two iOS devices (native app only — web is fine), but not on Android, Mac, or Windows.
Sadly, the error returned is not related to the cause.
Nothing seems too help.
The web page implies you can try it immediately. Initially it wasn't available.
A few hours later it was in both the web UI and the mobile app - I got a popu[ telling me that GPT-4o was available. However nothing seems to be any different. I'm not given any option to use video as an input, the app can't seem to pick up any new info from my voice.
I'm left a bit confused as to what I can do that I couldn't do before. I certainly can't seem to recreate much of the stuff from the announcement demos.
I imagine that there is a lot of usage at the HQ, human + AI karaoke?
It will be fully available in Eu with the GDPR compliance?
Consequences of audio2audio (rather than audio >text text>audio). Being able to manipulate speech nearly as well as it manipulates text is something else. This will be a revelation for language learning amongst other things. And you can interrupt it freely now!
I could be wrong but I haven't seen any non-speech demos.
Also it's possible this is trained on mostly speech.
It seems like we're in the skeuomorphism phase of AI where tools try to mimic humans like software tried mimic physical objects in the early 2000's.
I can't wait for us to be passed that phase.
"It looks like you entered the wrong number! Did you want to try again? Or did you want to talk to an agent?"
That sort of chirpy, overly enthusiastic voice?
I believe it can be toned down using system prompts, which they'll expose in future iterations
chuckling to 0%
no acting surprised
not making bullshit when you don't knowLLMs today have no concept of epistemology, they don't ever "know" and are always making up bullshit, which usually is more-or-less correct as a side effect of minimizing perplexity.
It worked surprisingly well. The video where he created the first iteration on stream(don't remember the watch party streams he ran the fans on): https://yewtu.be/watch?v=MBKouvwaru8
Also,
<spoilers>
It didn't do anything bad to Zaphod Beeblebrox, in a pocket universe created especially for him (therefore ensuring that he was the most important thing in it, and thereby securing his immunity from the mind-scrambling effects of fully comprehending the infinite smallness of one's place in the real universe).
(not that this is the most important thing about the announcement at all. Just an aside)
Yeh it's cringe. I had to stop listening.
Why did they make the woman sound like she's permanently on the brink of giggling? It's nauseating how overstated her pretentious banter is. Somewhere between condescending nanny and preschool teacher. Like how you might talk to a child who's at risk of crying so you dial up the positive reinforcement.
The new voice mode sounds better, but the current voice mode did also have inflection that made it feel much more natural than most computer voices I've heard before.
Being able to specifically request different tones is a new and very interesting feature.
Link in case other readers are curious: https://llm.datasette.io
Magic.
I also have an anecdote where it served (successfully) as a mediator for a couple.
Exciting times.
————
ChatGPT: Enjoy your breakfast and time together.
User: Can you tell that it's not just me talking to you right now?
ChatGPT: I can't always tell directly, but it sounds like you're sharing the conversation with someone else. Is [wife] there with you?
User: My god, the AI has awoken. Yes, this is [wife].
ChatGPT: Hi [wife]! It's great to hear from you. How are you doing?
User: I'm good. Thanks for asking. How are you?
ChatGPT: I'm doing well, thanks! How's everything going with the baby preparations?
—————
We were shocked. It was one of those times where it’s 25% heartwarming and 75% creepy. It was able to do this in part due to the new “memory” feature, that memorized my wife’s name and that we are expecting. it’s a strange novelty now, but this will be totally normalized and ubiquitous quite soon. Interesting times to be living in.
Based on the casual production of these videos, the product must be this good.
We've already seen how much damage dishonest actors can do by manipulating our text communications with words they don't mean, plans they don't intend to follow through on, and feelings they don't experience. The social media disinfo age has been bad enough.
Are you sure you want a machine which is able to manipulate our emotions on an even more granular and targetted level?
LLMs are still machines, designed and deployed by humans to perform a task. What will we miss if we anthropomorphize the product itself?
The only slightly annoying thing at the moment is they seem hard to interrupt, which is an important mechanism in conversations. But that seems like a solvable problem. They kind of need to be able to interpret body language a bit to spot when the speaker is about to interrupt.