Radar trends to watch: July 2022 – O’Reilly

This month, the big models are in the news more than last month: Bloom’s open source model is about to expire, Google’s lambda is good enough to trick people into thinking it’s conscious, and DALL-E just got better at drawing what you’re asking.

Perhaps the most important problem facing technology now is privacy protection. While this is not a new concern, it is one that most computer users would like to ignore, and that most tech companies have been willing to let them ignore. New state laws that criminalize out-of-state abortions and the storage of location information by anti-abortion groups have made privacy an issue that cannot be ignored.

Learn faster. I dig deeper. see further.

Artificial intelligence

• Big Science has almost finished his training open source BLOOM لغة Language ModelDeveloped by volunteer researchers and trained using public funds. Bloom will provide an open public platform for research into the capabilities of large linguistic models, and specifically, issues such as avoiding bias and toxic language.
• AI tools like AlphaFold2 . can Formation of new proteins, and not merely an analysis of the existing; The unexpected creation of new artifacts by an AI system is called “hallucination”. The proteins designed so far may not be useful. This remains a major step forward in drug design.
• Microsoft Restrict or remove access For some features in the face recognition service, Azure Face. Organizations will have to tell Microsoft how and why facial recognition will be used in their systems; Services such as emotion recognition will be removed entirely.
• Amazon plans to Alexa gives the ability to mimic anyone’s voice, using less than a minute of audio. They give the example of a grandmother (possibly dead) who “reads” a book to a child. Other AI vendors (most notably OpenAI / Microsoft) considered such a tradition immoral.
• idiot is an SQL database that allows you to release data using git commands, you can clone, push, drag, branch, fork and merge just like with git; You can access the data using SQL standard.
• It is unfortunately unsurprising that a robot that includes a widely used neural network (OpenAI CLIP) learns Racial and gender biasand that these biases affect his performance on the tasks.
• Building Self-driving vehicles with memory, so that they can recognize objects on the roads they are driving could be an important step in making AV practical. In real life, most people drive roads they are already familiar with. Autonomous vehicles should have the same feature.
• The debate continues over whether Google’s LaMDA is ‘conscious’, with a Google engineer placed on administrative leave for Post chat transcripts who claimed to be showing affection. Or just large linguistic models Squirrels?
• for Artists working in collaboration with Amnesty InternationalThe possibilities and drawbacks of artificial intelligence are a way to expand their creativity.
• Pete Warden’s suggestion ML . sensors It could make developing embedded ML systems simpler: Push machine learning into the sensors themselves.
• Researchers using DALL-E 2 found out The model has a “secret vocabulary” This is not human language, but it can be used fairly reliably to create consistent images. It may be the result of the model being unable to say “I didn’t get it”; Due to non-logical inputs, they are pulled towards similar words in the training set.
• Make HuggingFace File an agreement With Microsoft that will allow Azure customers Run HuggingFace Language Forms on Azure.
• startup Primary rule built Low code meta platform To build artificial intelligence systems. In the declarative system, you describe the result you want, rather than the process of creating the result. The system defines the process.
• Researchers develop artificial intelligence models that are implemented metamemory: The ability to remember whether you know something or not.
• As the population ages, it will be important to diagnose diseases such as Alzheimer’s early, when treatment can still be beneficial. AI Saving tools To help doctors analyze MRI images more accurately than humans. These tools do not attempt to diagnose. They provide data about the features of the brain.
• Google has Deepfakes training has been banned On Colab, the free Jupyter-based cloud programming platform.

metaverse

• Samsung and RedHat are working on New memory architectures and device drivers which would be sufficient for the demands of the cloud-based and 3D-enabled metaverse.
• The Metaverse Standards Forum It is a new industry group with the goal of solving Metaverse interoperability problems. It views the Metaverse as a product of the Web, and plans to coordinate work between existing sets of standards (eg the W3C) relevant to the Metaverse.
• can the “Open MetaverseTo be the future of the Internet? The Open the Metaverse Interoperability Group It builds vendor-independent benchmarks for social graphs, identities, and other elements of the Metaverse.
• 3D displays Allow 3D Augmented Reality: The ability to project 3D images onto the real world (for example, on the windshield of a car).
• The Google Visual Mode Service It uses the data they collected through Street View to provide high-accuracy positioning data to augmented reality applications. (This could be related to Niantic’s VPS server, or maybe they just use the same shortcut.)

protection

• With the end of Roe v. Wade, Personal data, including search records and location data, can be used to prosecute women who have had abortions. Data brokers already collect and sell this data. It’s unclear how large internet companies that collect this data will respond as well. (Google has announced that it will do so Delete location records which includes visits to sensitive websites.)
• Security researchers have identified more than 900,000 Kubernetes clusters exposed (and potentially vulnerable) to malicious scans. 65% of them are in the United States.
• Sonatype discovered a number of Modules in a PyPI repository in Python steal AWS credentials and other important data. Supply chain security will continue to be an issue for developers, regardless of the programming language or problem domain.
• Microsoft’s Analysis of Russia’s Cyber ​​War Efforts They showed that they increasingly attacked resources in countries allied with Ukraine (most notably the United States), and that government computers on the premises are particularly vulnerable.
• Working with Fastly and Cloudflare, Apple developed a service called automatic verification This eliminates the need for Captchas. According to rumors, it will be enabled by default in the iOS16 beta.
• Surprisingly small bot network (5000 hosts only) has produced a file Record DDOS Attack which peaked at 26 million HTTPS requests per second. The botnet was very powerful because most of its hardware belonged to cloud service providers. Cloudflare’s free service mitigated the attack.
• A different kind of attack against neural networks: introduce them Inputs that drive energy consumption in the worst caseThis forces processors to slow down or overheat.
• A new attack called Hertzbleed is used Small differences in processor clock speed While processing encryption keys to guess those keys. Intel and AMD processors are weak. While this attack may never be seen, it shows how the complexity of modern processors creates vulnerabilities.
• Symbiote is a new type of malware that attacks Linux, injects software into all running processes, and uses Berkeley packet filters (eBPF) to steal data and create secret communications channels. Symbiote uses Dynamic link hijacking To link executables to modified system libraries at runtime.
• In the first quarter of 2022, The number of known ransomware attacks decreased by 40%.This is largely due to the disappearance of the Conti ransomware suite. This decrease may only be temporary. Tactics have also changed. Attackers do not announce the names of their victims publicly, and prefer to negotiate ransoms privately.

Programming

• Amazon launched CodeWhispererwhich is a direct competitor to GitHub Copilot.
• Linus Torvalds expects it Rust will be used in the Linux kernel by 2023.
• GitHub Co-Pilot Now generally available (for the price); It is free for students and maintainers of open source software. Corporate licenses will be available later this year.
• WebAssembly is making progress. WebAssembly public runtime, WasmerRun any code on any platform. Impressive, if it delivers.
• Can WebAssembly replace Docker? Perhaps, in some applications. WASM provides portability and removes some security issues (maybe its own rendering); Docker sets up environments.
• Mozilla bergamot project It is a machine translation tool designed for use on the web. It can be used to build multilingual forms and other web pages. Unlike most other AI technologies, Bergamot works in the browser using WASM. No data is sent to the cloud.
• Microsoft has chest frame called liquid To build collaborative apps, such as Slack, Discord, and Teams. Microsoft will also release Azure Liquid Relay To support fluid-based applications.
• dragonfly It is a new in-memory database that claims much faster performance than memcached and Redis.
• The Chinese government has Block access to open source code on Gitee, the Chinese equivalent of GitHub, saying that all code must be reviewed by the government before it is released to the public.

web 3

• Is Blockchain Decentralized? Study commissioned by DARPA It investigates whether the blockchain is truly immutable, or whether it can be modified without exploiting weaknesses in cryptography, but by attacking the implementation protocols, networks, and consensus of the blockchain. This is the most comprehensive examination of blockchain security we have seen.
• Jack Dorsey announced that he is working on it web 5which will focus on identity management and is based on Bitcoin.
• Post Molly White questioning the possibility of An acceptably unfortunate sovereign identity You must read. It has an excellent summary and critique of all the work going on in this field.
• Encoder Matthew Green Is Important The Technologies Argument Behind Cryptocurrency (Although not for current applications).

biology

Quantitative Statistics

• Prospect computers, built from p-bits, may provide an important step forward for probabilistic decision making. This sounds esoteric, but it’s basically what we’re asking AI systems to do. P-bits may also be able to simulate q-bits and quantum computing.
• an system Connects two time crystals It could be the basis for a new form of quantum computing. Time crystals can exist at room temperature, and remain coherent for much longer than current qubit technologies.