These days, every phone has a camera, and few of us are ever without one. [Bjørn Karmann] has built an altogether not-camera, though, in the form of the Paragraphica, powered by artificial intelligence.
The Paragraphica doesn’t actually take photographs at all. Instead, it uses GPS to determine the user’s current position. It then feeds the address, time of day, weather, and temperature into a paragraph which serves as a prompt for an AI image generator. It also uses data gathered from various APIs to determine points of interest in the immediate area, and feeds those into the prompt as well. It then generates an artificial image that is intended to bear some resemblance to the prompt, and ideally, the real-world scene. In place of a lens, it bears a 3D printed structure inspired by the star-nosed mole, which feels its way around in lieu of using its eyes.
Three dials on the Paragraphica control its action. The first dial controls the radius of the area which the prompt will gather data about; it’s akin to setting the focal length of the lens. The second dial provides a noise seed value for the AI image generator, and the third dial controls how closely the AI sticks to the generated textual prompt.
The results are impressive, if completely false and generated from scratch. The Paragraphica generates semi-believable photos of a crowded alley, a public park, and a laneway full of parked cars. It’s akin to telling a friend where you are and what you’re seeing over the phone, and having them paint a picture based on that description.
Through their unique abilities and stolen data sets, AI image generators are proving controversial to say the least. As all good art does, Paragraphica explores this and raises new questions of its own.
AR and VR developer [Skarredghost] got pretty excited about a virtual blue cube, and for a very good reason. It marked a successful prototype of an augmented reality experience in which the logic underlying the cube as a virtual object was changed by AI in response to verbal direction by the user. Saying “make it blue” did indeed turn the cube blue! (After a little thinking time, of course.)
It didn’t stop there, of course, and the blue cube proof-of-concept led to a number of simple demos. The first shows off a row of cubes changing color from red to green in response to musical volume, then a bundle of cubes change size in response to microphone volume, and cubes even start moving around in space.
The program accepts spoken input from the user, converts it to text, sends it to a natural language AI model, which then creates the necessary modifications and loads it into the environment to make runtime changes in Unity. The workflow is a bit cumbersome and highlights many of the challenges involved, but it works and that’s pretty nifty.
The GitHub repository is here and a good demonstration video is embedded just under the page break. There’s also a video with a much more in-depth discussion of what’s going on and a frank exploration of the technical challenges.
If you’re interested in this direction, it seems [Skarredghost] has rounded up the relevant details. And should you have a prototype idea that isn’t necessarily AR or VR but would benefit from AI-assisted speech recognition that can run locally? This project has what you need.
Continue reading “Simple Cubes Show Off AI-Driven Runtime Changes In VR” →
Recently, an amusing anecdote made the news headlines pertaining to the use of ChatGPT by a lawyer. This all started when a Mr. Mata sued the airline where years prior he claims a metal serving cart struck his knee. When the airline filed a motion to dismiss the case on the basis of the statute of limitations, the plaintiff’s lawyer filed a submission in which he argued that the statute of limitations did not apply here due to circumstances established in prior cases, which he cited in the submission.
Unfortunately for the plaintiff’s lawyer, the defendant’s counsel pointed out that none of these cases could be found, leading to the judge requesting the plaintiff’s counsel to submit copies of these purported cases. Although the plaintiff’s counsel complied with this request, the response from the judge (full court order PDF) was a curt and rather irate response, pointing out that none of the cited cases were real, and that the purported case texts were bogus.
The defense that the plaintiff’s counsel appears to lean on is that ChatGPT ‘assisted’ in researching these submissions, and had assured the lawyer – Mr. Schwartz – that all of these cases were real. The lawyers trusted ChatGPT enough to allow it to write an affidavit that they submitted to the court. With Mr. Schwartz likely to be sanctioned for this performance, it should also be noted that this is hardly the first time that ChatGPT and kin have been involved in such mishaps.
Continue reading “ChatGPT V. The Legal System: Why Trusting ChatGPT Gets You Sanctioned” →
Researchers in Canada and the United States have used deep learning to derive an antibiotic that can attack a resistant microbe, acinetobacter baumannii, which can infect wounds and cause pneumonia. According to the BBC, a paper in Nature Chemical Biology describes how the researchers used training data that measured known drugs’ action on the tough bacteria. The learning algorithm then projected the effect of 6,680 compounds with no data on their effectiveness against the germ.
In an hour and a half, the program reduced the list to 240 promising candidates. Testing in the lab found that nine of these were effective and that one, now called abaucin, was extremely potent. While doing lab tests on 240 compounds sounds like a lot of work, it is better than testing nearly 6,700.
Interestingly, the new antibiotic seems only to be effective against the target microbe, which is a plus. It isn’t available for people yet and may not be for some time — drug testing being what it is. However, this is still a great example of how machine learning can augment human brainpower, letting scientists and others focus on what’s really important.
WHO identified acinetobacter baumannii as one of the major superbugs threatening the world, so a weapon against it would be very welcome. You can hope that this technique will drastically cut the time involved in developing new drugs. It also makes you wonder if there are other fields where AI techniques could cull out alternatives quickly, allowing humans to focus on the more promising candidates.
Want to catch up on machine learning algorithms? Google can help. Or dive into an even longer course.
AI image generators have gained new tools and techniques for not just creating pictures, but modifying them in consistent and sensible ways, and it seems that every week brings a fascinating new development in this area. One of the latest is Drag Your GAN, presented at SIGGRAPH 2023, and it’s pretty wild.
It provides a point-dragging interface that modifies images based on their implied structure. A picture is worth a thousand words, so this short animation shows what that means. There are plenty more where that came from at the project’s site, so take a few minutes to check it out.
GAN stands for generative adversarial network, a class of machine learning that features prominently in software like image generation; the “adversarial” part comes from the concept of networks pulling results between different goalposts. Drag Your GAN has a GitHub repository where code is expected to be released in June, but in the meantime, you can read the full paper or brush up on the basics of how AI image generators work, as well as see how image generation can be significantly enhanced with an understanding of a 2D image’s implied depth.
For a brief window of time in the mid-2010s, a fairly common joke was to send voice commands to Alexa or other assistant devices over video. Late-night hosts and others would purposefully attempt to activate voice assistants like these en masse and get them to do ridiculous things. This isn’t quite as common of a gag anymore and was relatively harmless unless the voice assistant was set up to do something like automatically place Amazon orders, but now that much more powerful AI tools are coming online we’re seeing that joke taken to its logical conclusion: prompt-injection attacks. Continue reading “Prompt Injection: An AI-Targeted Attack” →
We’re still in the early days of generatively-designed objects, but when combined with the capabilities of 3D printing, we’re already seeing some interesting results. One example is this new copper aerospike engine. [via Fabbaloo]
A collaboration between startups Hyperganic (generative AI CAD) and AMCM (additive manufacturing), this 800 mm long aerospike engine may be the most complicated 3D print yet. It continues the exciting work being done with 3D printing for aerospace applications. The complicated geometries of rocket nozzles of any type let additive manufacturing really shine, so the combination of generative algorithms and 3D printed nozzles could result in some big leaps in coming years.
Aerospikes are interesting as their geometry isn’t pressure dependent like more typical bell-shaped rocket nozzles meaning you only need one engine for your entire flight profile instead of the traditional switching mid-flight. A linear aerospike engine was one of the main selling points for the cancelled VentureStar Space Shuttle replacement.
This isn’t the only generative design headed to space, and we’ve covered a few projects if you’re interested in building your own 3D printed rocket nozzles or aerospike engines. Just make sure you get clearance from your local aviation regulator before your project goes to space!