\nThis week in AI: Big tech bets billions on machine learning tools<\/br>
\n2023-06-24 21:37:09<\/br><\/p>\n
Keeping up with an industry as fast-moving as\u00a0AI<\/a>\u00a0is a tall order. So until an AI can do it for you, here\u2019s a handy roundup of the last week\u2019s stories in the world of machine learning, along with notable research and experiments we didn\u2019t cover on their own.<\/p>\n If it wasn\u2019t obvious already, the competitive landscape in AI \u2014 particularly the subfield known as generative AI \u2014 is red-hot. And it\u2019s getting hotter. This week, Dropbox launched its first corporate venture fund, Dropbox Ventures<\/a>, which the company said would focus on startups building AI-powered products that \u201cshape the future of work.\u201d Not to be outdone, AWS debuted<\/a> a $100 million program to fund generative AI initiatives spearheaded by its partners and customers.<\/p>\n There\u2019s a lot of money being thrown around in the AI space, to be sure.\u00a0Salesforce Ventures, Salesforce\u2019s VC division,\u00a0plans<\/a>\u00a0to pour $500 million into startups developing generative AI technologies. Workday\u00a0recently<\/a> added $250 million to its existing VC fund specifically to back AI and machine learning startups. And Accenture and PwC have announced that they plan to invest $3 billion and $1 billion, respectively, in AI.<\/p>\n But one wonders whether money is the solution to the AI field\u2019s outstanding challenges.<\/p>\n In an enlightening panel during a Bloomberg conference<\/a> in San Francisco this week, Meredith Whittaker, the president of secure messaging app Signal, made the case that the tech underpinning some of today\u2019s buzziest AI apps is becoming dangerously opaque. She gave an example of someone who walks into a bank and asks for a loan.<\/p>\n That person can be denied for the loan and have \u201cno idea that there\u2019s a system in [the] back probably powered by some Microsoft API that determined, based on scraped social media, that I wasn\u2019t creditworthy,\u201d Whittaker said. \u201cI\u2019m never going to know [because] there\u2019s no mechanism for me to know this.\u201d<\/span><\/p>\n It\u2019s not capital that\u2019s the issue. Rather, it\u2019s the current power hierarchy, Whittaker says.<\/p>\n \u201cI\u2019ve been at the table for like, 15 years, 20 years. I\u2019ve been<\/em> at the table. Being at the table with no power is nothing,\u201d she continued.<\/p>\n Of course, achieving structural change is far tougher than scrounging around for cash \u2014 particularly when the structural change won\u2019t necessarily favor the powers that be. And Whittaker warns what might happen if there isn\u2019t enough pushback.<\/p>\n As progress in AI accelerates, the societal impacts also accelerate, and we\u2019ll continue heading down a \u201chype-filled road toward AI,\u201d she said, \u201cwhere that power is entrenched and naturalized under the guise of intelligence and we are surveilled to the point [of having] very, very little agency over our individual and collective lives.\u201d<\/p>\n That should<\/em> give the industry pause. Whether it actually will<\/em> is another matter. That\u2019s probably something that we\u2019ll hear discussed when she takes the stage at Disrupt in September.<\/p>\n Here are the other AI headlines of note from the past few days:<\/p>\n This week was CVPR up in Vancouver, Canada, and I wish I could have gone because the talks and papers look super interesting. If you can only watch one, check out Yejin Choi\u2019s keynote<\/a> about the possibilities, impossibilities, and paradoxes of AI.<\/p>\n Image Credits:<\/strong> CVPR\/YouTube<\/p>\n<\/div>\n The UW professor and MacArthur Genius grant recipient first addressed a few unexpected limitations of today\u2019s most capable models. In particular, GPT-4 is really bad at multiplication. It fails to find the product of two three-digit numbers correctly at a surprising rate, though with a little coaxing it can get it right 95% of the time. Why does it matter that a language model can\u2019t do math, you ask? Because the entire AI market right now is predicated on the idea that language models generalize well to lots of interesting tasks, including stuff like doing your taxes or accounting. Choi\u2019s point was that we should be looking for the limitations of AI and working inward, not vice versa, as it tells us more about their capabilities.<\/p>\n The other parts of her talk were equally interesting and thought-provoking. You can watch the whole thing here<\/a>.<\/p>\n Rod Brooks, introduced as a \u201cslayer of hype,\u201d gave an interesting history of some of the core concepts of machine learning<\/a> \u2014 concepts that only seem new because most people applying them weren\u2019t around when they were invented! Going back through the decades, he touches on McCulloch, Minsky, even Hebb \u2014 and shows how the ideas stayed relevant well beyond their time. It\u2019s a helpful reminder that machine learning is a field standing on the shoulders of giants going back to the postwar era.<\/p>\n Many, many papers were submitted to and presented at CVPR, and it\u2019s reductive to only look at the award winners, but this is a news roundup, not a comprehensive literature review. So here\u2019s what the judges at the conference thought was the most interesting:<\/p>\n Image Credits:<\/strong> AI2<\/p>\n<\/div>\n VISPROG<\/a>, from researchers at AI2, is a sort of meta-model that performs complex visual manipulation tasks using a multi-purpose code toolbox. Say you have a picture of a grizzly bear on some grass (as pictured) \u2014 you can tell it to just \u201creplace the bear with a polar bear on snow\u201d and it starts working. It identifies the parts of the image, separates them visually, searches for and finds or generates a suitable replacement, and stitches the whole thing back again intelligently, with no further prompting needed on the user\u2019s part. The Blade Runner \u201cenhance\u201d interface is starting to look downright pedestrian. And that\u2019s just one of its many capabilities.<\/p>\n \u201cPlanning-oriented autonomous driving,\u201d<\/a> from a multi-institutional Chinese research group, attempts to unify the various pieces of the rather piecemeal approach we\u2019ve taken to self-driving cars. Ordinarily there\u2019s a sort of stepwise process of \u201cperception, prediction, and planning,\u201d each of which might have a number of sub-tasks (like segmenting people, identifying obstacles, etc). Their model attempts to put all these in one model, kind of like the multi-modal models we see that can use text, audio, or images as input and output. Similarly this model simplifies in some ways the complex inter-dependencies of a modern autonomous driving stack.<\/p>\n \u00a0<\/p>\n DynIBaR<\/a> shows a high-quality and robust method of interacting with video using \u201cdynamic Neural Radiance Fields,\u201d or NeRFs. A deep understanding of the objects in the video allows for things like stabilization, dolly movements, and other things you generally don\u2019t expect to be possible once the video has already been recorded. Again\u2026 \u201cenhance.\u201d This is definitely the kind of thing that Apple hires you for, and then takes credit for at the next WWDC.<\/p>\n DreamBooth<\/a> you may remember from a little earlier this year when the project\u2019s page went live. It\u2019s the best system yet for, there\u2019s no way around saying it, making deepfakes. Of course it\u2019s valuable and powerful to do these kinds of image operations, not to mention fun, and researchers like those at Google are working to make it more seamless and realistic. Consequences\u2026 later, maybe.<\/p>\n The best student paper award goes to a method for comparing and matching meshes, or 3D point clouds \u2014 frankly it\u2019s too technical for me to try to explain, but this is an important capability for real world perception and improvements are welcome. Check out the paper here for examples and more info.<\/a><\/p>\n Just two more nuggets: Intel showed off this interesting model, LDM3D<\/a>, for generating 3D 360 imagery like virtual environments. So when you\u2019re in the metaverse and you say \u201cput us in an overgrown ruin in the jungle\u201d it just creates a fresh one on demand.<\/p>\n And Meta released a voice synthesis tool called Voicebox<\/a> that\u2019s super good at extracting features of voices and replicating them, even when the input isn\u2019t clean. Usually for voice replication you need a good amount and variety of clean voice recordings, but Voicebox does it better than many others, with less data (think like 2 seconds). Fortunately they\u2019re keeping this genie in the bottle for now. For those who think they might need their voice cloned, check out Acapela<\/a>.<\/p>\n<\/p><\/div>\n <\/br><\/br><\/br><\/p>\n\n
Other machine learnings<\/h2>\n
![\"\"](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/06\/choi-trans.jpg?resize=1024%2C473&ssl=1\"){kind=spacer}
<\/p>\n![\"\"](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/06\/imgedit_rationale.png?resize=1024%2C711&ssl=1\")
<\/p>\n![\"\"](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/06\/auton-planning.jpg?resize=1024%2C284&ssl=1\")
<\/p>\n