Hihi All. So I've just written this Patreon post <https://www.patreon.com/posts/104222987> all about OpenAi's latest release, ChatGPT-4o. I'm curious to see what peeps here make of it:
-- _ChatGPT-4o: Breakthrough or Bust?_ ChatGPT-4o [or Omni as the cool kids at OpenAI term it] sauntered into the release-spotlight earlier today, with OpenAi writing on their website that it’s: *“…a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, and image and generates any combination of text, audio, and image outputs. It can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in a conversation.”* Released alongside a bevy of slick videos, ChatGPT-4o is touted to be the next big thing with multiple video promos parading its new multimodal chops. One such video shows one version of GPT-4o narrating visual scenes to its visually impaired AI mate [while it asks questions in turn] all while the human testers fidgeted impatiently on the sidelines, barely masking their urge to fast-forward to the good bits. The grand unveiling of GPT-4o could’ve been lifted straight from a sci-fi script. We watched [some in awe, some with cynicism] as these videos attempted to paint a future where AIs chat about the décor in an empty room and then sing a duet about the process. And yet it's hard not to chuckle at the not-so-subtle desperation in the human testers who seem hell-bent on skipping/interrupting the more stilted voice-scripted parts of the AI dialogue and hurry to reach the 'money shot' of the demo [ie the concrete feature they were trying to showcase]. It begs the question: was the fanfare a bit premature? After all, this wasn’t the GPT-5 release party everyone had RSVP’d to. Diving into the nitty-gritty, the AI’s voice tech is frankly impressive, reminding us of those Figure 01 Speech-to-Speech Reasoning clips <https://www.youtube.com/watch?v=Sq1QZB5baNw&ab_channel=Figure> from a while back. The voices are spot-on: the AI sounds like a person, with natural human-emulated cadences, sub-vocalisations, and tonalities [the laughter is weirdly realistic]. What was far less impressive is the replication of bias in the gendered aspects the simulated speech with the female voices being far more sexualised/flirty than the male ones. It's especially disappointing given the potential ramifications for how this will impact users and perpetuate current gender biases. In 2024, one would hope we’d be past such gendered gimmicks [but….no]. Now let’s talk timing: releasing ChatGPT-4o for free might seem like a clever move, but the cynics among us might sniff something fishy. Rumour has it OpenAI’s nearly chewed through the entire web for data to feed its language models, so why not throw open the gates and let the global crowd serve up fresh fodder? It’s a clever ploy about scraping up every last crumb of human interaction to power their data-hungry tech. Let’s not be too harsh, though. OpenAI’s latest toy is a bit of a marvel: in some ways it’s like watching a new species come to life. But as we ooh and ahh over this latest OpenAI release, are we so dazzled by the prospect of talking chat [and vision-processing] bots that we overlook the less glamorous implications, like privacy erosion and data exploitation? [And let’s not kid ourselves, wasn't everyone really hanging out for GPT-5?] In essence ChatGPT-4o is a bit of a mixed bag. It's an impressive party trick, sure, but when the release-glitter settles we’re left pondering what it really brings to the table. It’s a step forward no doubt, but also a sidestep, a fancy detour on the road to more profound innovations. The real kicker isn’t what this AI can do but what it represents in the grand scheme of things [all up a blend of breakthrough and bias, of marvels and missed opportunities]. So let’s marvel at the spectacle but remain skeptical of the smoke and mirrors. After all [in the current world of current AI release-hype] every new release is a double-edged sword, and it’s up to us to suss out which edge to grab. -- | mezbreezedesign.com
_______________________________________________ NetBehaviour mailing list [email protected] https://lists.netbehaviour.org/mailman/listinfo/netbehaviour
