This is the regular weekly post where I share updates about things happening in gaming, machine learning, audio, or frankly anything that catches my eye and seems relevant to share or discuss here. Let’s hop right in!
New research out of MIT Lincoln Laboratory suggests that COVID-19 may be detectable through subtle changes to speech patterns, even for otherwise asymptomatic speakers. If true, this would open up an extremely powerful and non-invasive way to quickly and regularly screen folks who may have been exposed to COVID. That said, it’s a bit early to celebrate. For one thing, the researchers themselves confess the research is still early – and while I’d be overjoyed to be wrong here, my prior that this can be done is pretty low. Our voices are highly variable and diverse, so I find it unlikely there would be a unique signal of “has coronavirus” which never ever appears in one’s voice for any other reason. In which case, the question will be whether things like taking a baseline reading when the speaker is healthy will be enough to reduce false positives to an acceptable level. Worth keeping an eye on, at least!
A new frontrunner for the Guinness Book of Unreasonably Large Neural Networks, OpenAI’s GPT-3, which is the next generation of its text generation algorithm, used 175 billion parameters and cost $12m to train at all. What’s truly powerful about GPT-3 is its ability to operate based on extremely few prior examples of a given text problem, resulting in not only compelling Chinese-Room-style shows of intelligence but opportunities to apply it to nearby, but different, problem spaces like generating proper layout code for a website based only on a verbal description of the page. Now, before anyone freaks out too much, GPT-3 does write human-sounding news articles, but it’s still difficult if not impossible to get it to actually write about the specific, actual news events you want it to (instead of writing plausible-sounding accounts of fictional events.) And the sheer scale of the network decreases the usability of it, meaning it’s going to be tough to iterate on it and solve the hard problems that separate what’s effectively a very compelling magic trick from actually being able to solve real important problems. Nevertheless, though, this is an amazing leap forward, and I’m sure the potential of this network is still largely untapped.
The Generalist discusses what they see as the next phase of dating apps – now that we’ve abstracted away things like location (by connecting people digitally) and appearance (by removing or changing the way we engage with images of each other), they anticipate the next generation abstracting away human partners entirely. Whether or not you agree with (or are excited about) the premise, the article does a good job outlining the various elements that have led people to develop closer and closer bonds with digital beings in the past few years – including a whopping 40% of Replika’s 500K users who report romantic feelings for their virtual companions. Of course, the same technology which might enable romantic relationships could also drive virtual celebrities, simple friendships, or perhaps even professional tools such as a digital, trusted avatar which can always provide a listening ear and offers a first line of therapy for those who need it.
Google Stadia is still charging forward, recently announcing partnerships with Harmonix as well as Supermassive Games. Harmonix in particular strikes me as an interesting choice – rhythm games managed on a remote server seem like a high risk for latency to cause significant problems, but I suppose if they can make it happen it will be a great proof of concept for Stadia to show that they can keep that latency down.
And speaking of game streaming, Microsoft’s xCloud is coming in September (though it may have a new name by then.) It will initially only be available to Xbox Game Pass Ultimate subscribers, though they promise it will become more available over time. More news should be coming in August, which will hopefully shed some valuable light on how Microsoft and Xbox see game streaming within their larger ambitions moving forward.
The Trump administration reversed course, cancelling the threat of deportation for students here on visas who might be unable to attend in-person classes next semester due to COVID. (In particular, this happened shortly after several prominent universities, including Harvard and MIT, began to draft lawsuits against the original decision.) I was extremely pleased to see this reversion – even without getting into politics around immigration in general, it’s pretty easy to argue that “extremely brilliant people who wish to come here to invent and create value within America, and for Americans” is a particularly important group of folks to continue to welcome here!
Twitter faced a massive breach on Wednesday, in which hackers used social engineering to gain access to an internal admin tool, and in turn compromised several prominent accounts. The hack became obvious when the hackers began posting messages attempting to scam people into sending them bitcoin via the Twitter accounts of individuals like Elon Musk, Jeff Bezos, Joe Biden, and Barack Obama; but it remains unclear if this is all the hackers were after, or if this was a smokescreen or signal meant to facilitate a more insidious purpose. Twitter has blogged about the incident but is still attempting to assess the extent of the damage for a few accounts, so this story is definitely still evolving. Regardless of the actual damage this time, though, I’m sure I’m not the only one to suggest that the fact this could happen in the first place should be kept in mind for anyone who uses Twitter prominently, and especially those who rely on it functionally for private discussions or crucial messaging.
That’s all for this week! As always, any thoughts or feedback are welcome. Stay safe, healthy, and sane, all!