3 ms·

You're reading my post too optimistically. I am far from convinced that computing is a good volcano to leap into these days. The last innovations that have made

You're reading my post too optimistically. I am far from convinced that computing is a good volcano to leap into these days. The last innovations that have made me wow were the Wikipedia (2001), python (1991-) and mid-2000s progress in game engines. Over the last 15 years, all the changes have been rather incremental (languages, security, some OSs, game engines again) or disappointing (cryptocurrencies, social networks), with the exception of ML. Now, ML is a fresh breeze but it is lacking the "garage-ability" that computing has been famous for; good luck competing with the bigshots on that field. And there is still a great potential for disappointment as it is unclear how well it scales beyond the type of problems that have lots of data to train on or allow for automatic reward (RL).

jodrellblank 4y ago
I like your canary; it reminds me of the Terence McKenna quote "The artist's task is to save the soul of mankind; and anything less is a dithering while Rome burns. Because if the artists - who are self-selected for being able to journey into the Other - if the artists cannot find the way, then the way cannot be found.". If the hackers - who are self-selected for being able to journey into the Machine - cannot find the adventure, then the adventure cannot be found. Things which have made me wow in more recent years: - Large scale aggregation of data, specifically when Google started using data from phones around the planet to overlay live road traffic data on Maps, or when individual businesses are typically busy on different days of the week. Live map views of lightning ( https://lightningmaps.org/ https://lightningmaps.org/ ) or weather ( https://www.windy.com/ https://www.windy.com/ ). Things a skilled programmer might build (a map on a website) but which can't work without a global network of sensors. - More continuously active sensors, brought about by specialist circuitry and low power systems, e.g. a phone with raise to wake, step tracker, and which silence incoming calls when they detect unusually stressful manouevres in a car. - JSLinux and v86 browser-based virtual machines. I know the tech of running a VM isn't particularly knew but the ability to boot a Linux/Windows VM with one click in a browser almost everyone has without needing a cloud/container instance behind it feels like it will bloom into a lot of new things over time. - The first app to run on smartphones which did OCR of the camera, translation of the text, and live overlay of the translated text back on the picture on screen. I forget its name now, and now it's builtin to cameraphones. - Going back about 15 years, but when content aware scaling came in - https://en.wikipedia.org/wiki/Seam_carving https://en.wikipedia.org/wiki/Seam_carving - Something else which is scale related but not ML, when a cloud storage program like DropBox can hash a file on your local machine, send the hash to their servers, notice that someone else has already uploaded that file, and tag it into your account so you can 'upload' non-unique files without the time or bandwidth of actually uploading them. - When DropBox live recompress JPGs using a lossless compression to save tens of petabytes of storage, then decompress them back into JPG as people access them. https://blog.acolyer.org/2017/05/01/the-design-implementation-and-deployment-of-a-system-to-transparently-compress-hundreds-of-petabytes-of-image-files-for-a-file-storage-service/ https://blog.acolyer.org/2017/05/01/the-design-implementatio... - Internet traffic highjacking by 'bitsquatting' domain names which are one memory-bit-flipped-bit away from the correct name, and then using the incoming traffic to estimate the global amount of flakey memory and cosmic ray events happening: https://www.google.com/search?hl=en&q=squatting%20memory%20bit%20flip%20domains https://www.google.com/search?hl=en&q=squatting%20memory%20b... - Deepfakes; they appeared as hype and then faded from hype, but the ability to synthesize another person's facial appearanace and vocal mannerisms on someone else is impressive. - Drone FPS flying with VR headsets. Coordinated light-drone displays in the sky instead of fireworks. - ML related, but iPhone accessibility features can describe pictures, live while using the camera ( https://youtu.be/UnoeaUpHKxY?t=39 https://youtu.be/UnoeaUpHKxY?t=39 ) or in the photos app or on websites. Or apps like Audible Vision (e.g. https://youtu.be/QiEKMTTwTZg?t=377 https://youtu.be/QiEKMTTwTZg?t=377 ). - ML related, Stable Diffusion, being able to generate visual scenes from text descriptions. > "Now, ML is a fresh breeze but it is lacking the "garage-ability" that computing has been famous for; good luck competing with the bigshots on that field." Doug Miles is trying with LogicMoo ( https://www.youtube.com/watch?v=sdG6GVCwJrw https://www.youtube.com/watch?v=sdG6GVCwJrw ), trying to build an AI that learns inside a virtual world instead of using big data and ML techniques.
- generationP 4y ago
  Never heard of seam carving before. Not sure if it's used anywhere, but what a fun idea it is! Lepton is interesting (as is brotli), although it doesn't quite rise to the level of "opening gates to new worlds" that the innovations around 2000 did. Virtual machines were funny in that they implemented an idea from mid-century mathematical logic, but VMWare was founded in 1998 and cygwin (not quite virtual machines but similar) is even older. Oh yeah, lots of innovation in spam, social engineering and trolling, but that's not computing per se :) Well aware of the march of big data and "quantity becoming quality"-based services over the last 20 years. But Google Earth and Wikipedia started in 2001, and everything that came thereafter would mostly be less exciting and more closed-down. OpenStreetMap deserves a mention, not for innovation but for stealing the fire from the gods. Windy.com is a fresh breeze, too; good point. Never found DropBox exciting. Even git, which I love and use for 3 different purposes every day, just doesn't feel particularly novel. Maybe that's because torrenting (with all its deduplication, hash-indexing and various other innovations) had set my expectations so high long ago that everything that came after looked like the Dark Ages. Drones... now these are some new grounds. In hindsight, I feel stupid forgetting them in my comment above!
adamsmith143 4y ago
>ML is a fresh breeze but it is lacking the "garage-ability" that computing has been famous for; good luck competing with the bigshots on that field. Not sure about this. Groups like Eleuther.ai or Stability have put out serious competitors to something like GPT-3 and Dall-e respectively without the massive capital that OpenAI has. HuggingFace also gives you access to virtually all the most cutting edge models and all you need to do is plug and play.

jodrellblank 4y ago

I like your canary; it reminds me of the Terence McKenna quote "The artist's task is to save the soul of mankind; and anything less is a dithering while Rome burns. Because if the artists - who are self-selected for being able to journey into the Other - if the artists cannot find the way, then the way cannot be found.". If the hackers - who are self-selected for being able to journey into the Machine - cannot find the adventure, then the adventure cannot be found. Things which have made me wow in more recent years: - Large scale aggregation of data, specifically when Google started using data from phones around the planet to overlay live road traffic data on Maps, or when individual businesses are typically busy on different days of the week. Live map views of lightning ( https://lightningmaps.org/ https://lightningmaps.org/ ) or weather ( https://www.windy.com/ https://www.windy.com/ ). Things a skilled programmer might build (a map on a website) but which can't work without a global network of sensors. - More continuously active sensors, brought about by specialist circuitry and low power systems, e.g. a phone with raise to wake, step tracker, and which silence incoming calls when they detect unusually stressful manouevres in a car. - JSLinux and v86 browser-based virtual machines. I know the tech of running a VM isn't particularly knew but the ability to boot a Linux/Windows VM with one click in a browser almost everyone has without needing a cloud/container instance behind it feels like it will bloom into a lot of new things over time. - The first app to run on smartphones which did OCR of the camera, translation of the text, and live overlay of the translated text back on the picture on screen. I forget its name now, and now it's builtin to cameraphones. - Going back about 15 years, but when content aware scaling came in - https://en.wikipedia.org/wiki/Seam_carving https://en.wikipedia.org/wiki/Seam_carving - Something else which is scale related but not ML, when a cloud storage program like DropBox can hash a file on your local machine, send the hash to their servers, notice that someone else has already uploaded that file, and tag it into your account so you can 'upload' non-unique files without the time or bandwidth of actually uploading them. - When DropBox live recompress JPGs using a lossless compression to save tens of petabytes of storage, then decompress them back into JPG as people access them. https://blog.acolyer.org/2017/05/01/the-design-implementation-and-deployment-of-a-system-to-transparently-compress-hundreds-of-petabytes-of-image-files-for-a-file-storage-service/ https://blog.acolyer.org/2017/05/01/the-design-implementatio... - Internet traffic highjacking by 'bitsquatting' domain names which are one memory-bit-flipped-bit away from the correct name, and then using the incoming traffic to estimate the global amount of flakey memory and cosmic ray events happening: https://www.google.com/search?hl=en&q=squatting%20memory%20bit%20flip%20domains https://www.google.com/search?hl=en&q=squatting%20memory%20b... - Deepfakes; they appeared as hype and then faded from hype, but the ability to synthesize another person's facial appearanace and vocal mannerisms on someone else is impressive. - Drone FPS flying with VR headsets. Coordinated light-drone displays in the sky instead of fireworks. - ML related, but iPhone accessibility features can describe pictures, live while using the camera ( https://youtu.be/UnoeaUpHKxY?t=39 https://youtu.be/UnoeaUpHKxY?t=39 ) or in the photos app or on websites. Or apps like Audible Vision (e.g. https://youtu.be/QiEKMTTwTZg?t=377 https://youtu.be/QiEKMTTwTZg?t=377 ). - ML related, Stable Diffusion, being able to generate visual scenes from text descriptions. > "Now, ML is a fresh breeze but it is lacking the "garage-ability" that computing has been famous for; good luck competing with the bigshots on that field." Doug Miles is trying with LogicMoo ( https://www.youtube.com/watch?v=sdG6GVCwJrw https://www.youtube.com/watch?v=sdG6GVCwJrw ), trying to build an AI that learns inside a virtual world instead of using big data and ML techniques.

generationP 4y ago
Never heard of seam carving before. Not sure if it's used anywhere, but what a fun idea it is! Lepton is interesting (as is brotli), although it doesn't quite rise to the level of "opening gates to new worlds" that the innovations around 2000 did. Virtual machines were funny in that they implemented an idea from mid-century mathematical logic, but VMWare was founded in 1998 and cygwin (not quite virtual machines but similar) is even older. Oh yeah, lots of innovation in spam, social engineering and trolling, but that's not computing per se :) Well aware of the march of big data and "quantity becoming quality"-based services over the last 20 years. But Google Earth and Wikipedia started in 2001, and everything that came thereafter would mostly be less exciting and more closed-down. OpenStreetMap deserves a mention, not for innovation but for stealing the fire from the gods. Windy.com is a fresh breeze, too; good point. Never found DropBox exciting. Even git, which I love and use for 3 different purposes every day, just doesn't feel particularly novel. Maybe that's because torrenting (with all its deduplication, hash-indexing and various other innovations) had set my expectations so high long ago that everything that came after looked like the Dark Ages. Drones... now these are some new grounds. In hindsight, I feel stupid forgetting them in my comment above!

adamsmith143 4y ago

>ML is a fresh breeze but it is lacking the "garage-ability" that computing has been famous for; good luck competing with the bigshots on that field. Not sure about this. Groups like Eleuther.ai or Stability have put out serious competitors to something like GPT-3 and Dall-e respectively without the massive capital that OpenAI has. HuggingFace also gives you access to virtually all the most cutting edge models and all you need to do is plug and play.