Waydroid is "A container-based approach to boot a full Android system on a regular GNU/Linux system like Ubuntu."
As the name implies, it's built using Wayland, although I don't know in what way it depends on Wayland. Wayland is a replacement for the X11 window system.
"Rust is the fastest-growing programming language, with its developer community doubling in size over the past two years, yet JavaScript remains the most popular language with 25.2 million active developers, according to the results of a recent survey."
"Python has overtaken Java as the second most popular language, driven by the interest in machine learning and AI."
"Meanwhile, the Go language saw its developer population grow by 10% over the last year."
"Objective-C has stagnated for the last two years."
"Swift has seen a small growth over the past 12 months (5%) to 4.6 million developers, which led to it being overtaken by Go."
"Scholars: AI isn't 'hallucinating' -- it's bullshitting."
"To better understand why these inaccuracies might be better described as bullshit, it is helpful to look at the concept of bullshit as defined by philosopher Harry Frankfurt. In his seminal work, Frankfurt distinguishes bullshit from lying. A liar, according to Frankfurt, knows the truth but deliberately chooses to say something false. In contrast, a bullshitter is indifferent to the truth. The bullshitter's primary concern is not whether what they are saying is true or false but whether it serves their purpose, often to impress or persuade."
"Frankfurt's concept highlights that bullshit is characterized by a disregard for the truth. The bullshitter does not care about the accuracy of their statements, only that they appear convincing or fit a particular narrative."
Does this mean we can use the word "bullshitting" in polite conversation now?
Llamanet claims to be "an open source, drop-in library/tool that lets you instantly turn any OpenAI-powered apps into llama.cpp apps, with just one line."
"It works with zero setup and zero 3rd party dependency so you don't need to know anything, and just treat it like OpenAI."
"Llamanet is an embeddable llama.cpp management system that 'just works', automagically. It lets you instantly replace OpenAI with one line of code in your app. Because the engine is embedded in your app, you don't need to tell your users to install a 3rd party LLM app or server just to use your app."
Hmm. Wonder if the results quality is as good, though? Might be worth trying.
Resume.Ink claims that if you upload your résumé and a job description, it will make a version of your résumé tailored to that particular job description.
The world's biggest digital camera is 3.2 gigapixels. It weighs nearly three tons. And it was created for astronomy. It was installed in the Vera C. Rubin Observatory on the top of a mountain called Cerro Pachón in the Coquimbo region in northern Chile, on the edge of the Atacama desert.
https://gdb.voanews.com/b135c51e-f9a6-45d4-b11d-ecd7f3306330_cx0_cy10_cw0_w1200_r1.jpg
"Physicians argue in their amicus brief filed this week with the United States Ninth Circuit Court of Appeals that when Apple's watchOS 5 replaced its Heart Rate Path Optimizer (HRPO) algorithm with a Heart Rate Neural Network (HRNN) and Irregular Rhythm Notifications (IRN), it made the wearables worse for monitoring heart health."
Choice quotes I pulled from the their brief (24-page document):
"AliveCor, Inc. developed a heart rhythm analysis application -- Kardia -- which included a feature called SmartRhythm that patients could use with Apple, Inc.'s Apple Watch. SmartRhythm allowed patients to continuously monitor their own heart rhythms and receive notifications of irregularities in real time (every five seconds). Patients could then immediately utilize AliveCor's band to conduct an electrocardiogram ('ECG') if they were alerted to an irregular heart rhythm. This was a major breakthrough in technology and many of the undersigned recommended SmartRhythm for their own patients suffering from atrial fibrillation ('Afib') -- a condition that causes an irregular heart rhythm. Unfortunately, Apple introduced a new version of its Apple Watch ('watchOS 5') that cut off access to the raw heart data needed to continuously monitor heart rhythm and provide patients with real-time information. As a result, AliveCor had to pull SmartRhythm from the market."
"Amici curiae understand that Apple, in an attempt to defend its actions, points to the introduction of Irregular Rhythm Notification ('IRN') with watchOS 5 as a supposed product improvement. From a medical point of view, IRN is inferior when it comes to medical monitoring. IRN only sporadically measures a user's heart rhythm; and critically, unlike the AliveCor product, Apple's feature is not FDA cleared for users with Afib. Indeed, Apple itself advises Afib patients not to use its replacement product for heart rhythm monitoring."
"When Apple introduced the Apple Watch it provided third-party developers, including AliveCor, with access to continuous heart rhythms of Apple Watch users. This was accomplished using a key algorithm -- the Heart Rate Path Optimizer ('HRPO') -- in Apple's Workout Mode application program interface ('API'). Utilizing this HRPO algorithm, AliveCor's SmartRhythm feature within its Kardia application ('SmartRhythm') was able to detect heart rhythm irregularities and offered continuous real-time monitoring and analysis of a patient's heart rhythms. Importantly, the patient could be at rest or in motion and AliveCor's app would still monitor and provide alerts regarding irregular heart rhythms. This continuous monitoring was extremely valuable to patients suffering from Afib because a patient is at greater risk of a stroke with each Afib event. SmartRhythm also allowed patients to track the number of heart rhythm irregularities after a cariologist recommended certain medications or lifestyle changes, which can assist doctors in making adjustments and providing better individualized care."
"When Apple introduced watchOS 5, it introduced a new algorithm to the Workout Mode API called the Heart Rate Neural Network ('HRNN') algorithm. Apple then launched its own Irregular Rhythm Notification ('IRN') feature in watchOS 5. Apple's IRN feature only analyzes a patient's heart rhythms when they are at rest. After introducing watchOS 5, Apple stopped making HRPO-generated heart data available to third-party developers, like AliveCor."
"A key reason for AliveCor's decision to pull SmartRhythm from the market was that Apple's HRNN algorithm artificially modified ('smoothed') reported patient heart rate data. For understandable reasons of patient safety, AliveCore could not rely on heart rate data that Apple had elected to modify."
"The district court concluded, as a matter of law, that Apple's HRNN algorithm was an improvement over the HRPO algorithm for exercise purposes." "However, even under the district court's framework, it is undisputed that the HRNN algorithm is not an improvement for medical monitoring purposes."
Apple crippled watchOS to corner heart-tracking market, doctors say
Melancholia in the San Francisco Bay Area, at least that is Scott Sumner's experience.
"During my recent trip to the Bay Area, I met lots of people who are involved in the field of AI. My general impression is that this region has more smart people than anywhere else, at least per capita. And not just fairly smart, I'm talking about extremely high IQ individuals. I don't claim to have met a representative cross section of AI people, however, so take the following with a grain of salt."
"If you spend a fair bit of time surrounded by people in this sector, you begin to think that San Francisco is the only city that matters; everywhere else is just a backwater. There's a sense that the world we live in today will soon come to an end, replaced by either a better world or human extinction. It's the Bay Area's world, we just live in it."
"In other words, I don't know if the world is going to end, but it seems as though this world is coming to an end."
#solidstatelife #ai #technologicalunemployment #existentialrisk
"First NHS physiotherapy clinic run by AI to start this year."
Um, what does "run by AI" mean?
"The new platform will provide same-day automated video appointments with a digital physiotherapist via an app that responds to information provided by a patient in real time."
Alrighty then.
"PiWings 2.0 is a tiny drone based on Raspberry Pi Pico and ESP8266 WiFi module".
For those of you who are microcontroller enthusiasts and/or drone enthusiasts.
PiWings 2.0 is a tiny drone based on Raspberry Pi Pico and ESP8266 WiFi module
"blendOS v4 released: Arch Linux, made immutable, declarative and atomic."
"We're thrilled to unveil blendOS v4, a groundbreaking release that redefines blendOS as a highly-flexible, immutable and atomic variant of Arch Linux."
"A single file, /system.yaml, is used to define everything on your system."
No mention of NixOS, but this sounds like the NixOS philosophy, applied to the world of Arch. Now that the NixOS community has descended into infighting, maybe the door is open for blendOS and Arch Linux to really take off?
blendOS v4 released: Arch Linux, made immutable, declarative and atomic
"View and explore atomic models in Augmented Reality!"
I don't have a VR headset and I checked and my Android device does not support ARCore, so this is for those of you who can do VR.
"AR Atom Visualizer is an app that allows you to view and explore atomic models in Augmented Reality with Google ARCore on your smartphone. Many of us understand the basic structure of an atom: a nucleus containing protons and neutrons, surrounded by electrons - but how are those electrons organized? How do they move? What do they look like?"
"The Bohr model presents the atom as a shell containing a nucleus and the electrons in orbit around it. It helps us understand the energy level of electrons and how they are organised in relation to the nucleus."
"The quantum mechanical model presents the atom as an electron cloud. This helps us understand the possible location of the electrons in relation to the nucleus."
"AR Atom Visualizer uses Augmented Reality to create 3D animated visualizations of both these models of any atom in real space, just by using the camera on your smartphone."
AR Atom Visualizer - Growing STEM talent in the Signal Garden
sqlite-vec an SQLite extension for vector search under development.
"I'm working on a new SQLite extension! It's called sqlite-vec, an extension for vector search, written purely in C. It's meant to replace sqlite-vss, another vector search SQLite extension I released in February 2023, which has a number of problems. I believe the approach I'm taking with sqlite-vec solves a number of problem it's predecessor has, will have a much nicer and performant SQL API, and is a better fit for all applications who want an embed vector search solution!"
"sqlite-vec will be a SQLite extension written purely in C with no dependencies. It will provide custom SQL functions and virtual tables for fast vector search, as well as other tools and utilities for working with vectors (quantization, JSON/BLOB/numpy conversions, vector arithmetic, etc.)."
The full text of Dimitri P. Bertsekas's book A Course in Reinforcement Learning is available online for free. It's also available for purchase in print form. About 450 pages. It's the textbook for his course at Arizona State University "Reinforcement Learning and Optimal Control".
I've gone through more than half of Richard Sutton and Andrew Barto's book Reinforcement Learning: An Introduction (though I confess to have 'cheated' and not done all the exercises). It might be worth reading this book, too, to see the same material from an alternate point of view.
"Reinforcement learning can be viewed as the art and science of sequential decision making for large and difficult problems, often in the presence of imprecisely known and changing environment conditions. Dynamic programming is a broad and well-established algorithmic methodology for making optimal sequential decisions, and is the theoretical foundation upon which reinforcement learning rests. This is unlikely to change in the future, despite the rapid pace of technological innovation. In fact, there are strong connections between sequential decision making and the new wave of technological change, generative technology, transformers, GPT applications, and natural language processing ideas, as we will aim to show in this book."
"In dynamic programming there are two principal objects to compute: the optimal value function that provides the optimal cost that can be attained starting from any given initial state, and the optimal policy that provides the optimal decision to apply at any given state and time. Unfortunately, the exact application of dynamic programming runs into formidable computational difficulties, commonly referred to as the curse of dimensionality. To address these, reinforcement learning aims to approximate the optimal value function and policy, by using manageable off-line and/or on-line computation, which often involves neural networks (hence the alternative name Neuro-Dynamic Programming)."
"Thus there are two major methodological approaches in reinforcement learning: approximation in value space, where we approximate in some way the optimal value function, and approximation in policy space, whereby we construct a suboptimal policy by using some form of optimization over a suitably restricted class of policies."
"The book focuses primarily on approximation in value space, with limited coverage of approximation in policy space. However, it is structured so that it can be easily supplemented by an instructor who wishes to go into approximation in policy space in greater detail, using any of a number of available sources."
"An important part of our line of development is a new conceptual framework, which aims to bridge the gaps between the artificial intelligence, control theory, and operations research views of our subject. This framework, the focus of the author's recent monograph 'Lessons from AlphaZero ...',, centers on approximate forms of dynamic programming that are inspired by some of the major successes of reinforcement learning involving games. Primary examples are the recent (2017) AlphaZero program (which plays chess), and the similarly structured and earlier (1990s) TD-Gammon program (which plays backgammon)."
The full text of Simone Scardapane's book Alice's Adventures in a Differentiable Wonderland is available online for free. It's not available in print form because it's being written and this is actually a draft. But it looks like Volume 1 is pretty much done. It's about 260 pages. It introduces mathematical fundamentals and then explains automatic differentiation. From there it applies the concept to convolutional layers, graph layers, and transformer models. A volume 2 is planned with fine-tuning, density estimation, generative modeling, mixture-of-experts, early exits, self-supervised learning, debugging, and other topics.
"Looking at modern neural networks, their essential characteristic is being composed by differentiable blocks: for this reason, in this book I prefer the term differentiable models when feasible. Viewing neural networks as differentiable models leads directly to the wider topic of differentiable programming, an emerging discipline that blends computer science and optimization to study differentiable computer programs more broadly."
"As we travel through this land of differentiable models, we are also traveling through history: the basic concepts of numerical optimization of linear models by gradient descent (covered in Chapter 4) were known since at least the XIX century; so-called 'fully-connected networks' in the form we use later on can be dated back to the 1980s; convolutional models were known and used already at the end of the 90s. However, it took many decades to have sufficient data and power to realize how well they can perform given enough data and enough parameters."
"Gather round, friends: it's time for our beloved Alice's adventures in a differentiable wonderland!"
Alice's Adventures in a differentiable wonderland
#solidstatelife #aieducation #differentiation #neuralnetworks
The full text of Simon J.D. Prince's book Understanding Deep Learning is available online for free -- though the author asks that you buy the book and write a (positive, one would hope) review on Amazon. He will make a 2nd edition if sales are good.
The book is around 500 pages and a glance at the table of contents shows it goes from fundamentals to very advanced topics: Supervised learning, shallow neural networks, deep neural networks, loss functions (maximum likelihood, univariate regression, classification, cross-entropy, etc), gradient descent, stochastic gradient descent, initialization, the backpropagation algorithm, hyperparameters, regularization, convolutional neural networks, residual networks, transformers, graph neural networks, unsupervised learning, generative adversarial networks (styleGAN, etc), normalizing flows, variational autoencoders, diffusion models, reinforcement learning, why does deep learning work? and ethics. Appendices for notation, mathematics, and probability.
Simon J.D. Prince: Understanding Deep Learning
#solidstatelife #ai #deeplearning #sgd #backpropagation #genai #gans #aieducation
PyTorch documentary. This documentary isn't about the technology, it's about the people behind PyTorch. Meet the original creators like Adam Paszke, Soumith Chintala, and Yangqing Jia. Meet early adopters like Jeremy Howard. Meet people who brought PyTorch from a research tool to a production deployment like Lin Qiao. PyTorch is the invisible tool behind ChatGPT, Stable Diffusion, and many other AI products.
Official PyTorch Documentary: Powering the AI Revolution - PyTorch
Leaders at Humane, the "AI Pin" company, allegedly prohibited criticism within the company, and this may have played a role in their launch of a "dead-on-arrival" product.
"The Times interviewed '23 current and former employees, advisers and investors,' and their anecdotes shed a lot of light on how a company makes it all the way to market with an out-of-touch, poorly performing product. The two founders apparently 'preferred positivity over criticism, leading them to disregard warnings about the AI Pin's poor battery life and power consumption. A senior software engineer was dismissed after raising questions about the product, they said, while others left out of frustration.' After that software engineer was fired for questioning if the AI pin would be ready for launch, the report describes a staff meeting where the founders 'said the employee had violated policy by talking negatively about Humane.' It's hard to make a good product if you can't honestly talk about the negatives and positives for fear of retaliation."
How to build a DOA product: Humane AI Pin founders banned internal criticism
Somehow the "system prompt" for "artifacts" (for example those code sections) from Claude 3.5 Sonnet somehow got leaked.
These peaks behind the curtain are always interesting. The prompts are always bigger, more complicated, and stranger than you'd expect. Well, I don't know that this one is that strange but it's definitely big and complicated.
Goldman Sachs' take on AI: "Gen AI: Too much spend, too little benefit?"
"Tech giants and beyond are set to spend over $1tn on AI capex in coming years, with so far little to show for it. So, will this large spend ever pay off? MIT's Daron Acemoglu and Goldman Sachs' Jim Covello are skeptical, with Acemoglu seeing only limited US economic upside from AI over the next decade and Covello arguing that the technology isn't designed to solve the complex problems that would justify the costs, which may not decline as many expect. But Goldman Sachs' Joseph Briggs, Kash Rangan, and Eric Sheridan remain more optimistic about AI's economic potential and its ability to ultimately generate returns beyond the current 'picks and shovels' phase, even if AI's 'killer application' has yet to emerge. And even if it does, we explore whether the current chips shortage (with Goldman Sachs' Toshiya Hari) and looming power shortage (with Cloverleaf Infrastructure's Brian Janous) will constrain AI growth. But despite these concerns and constraints, we still see room for the AI theme to run, either because AI starts to deliver on its promise, or because bubbles take a long time to burst. Generative AI has the potential to fundamentally change the process of scientific discovery, research and development, innovation, new product and material testing, etc. as well as create new products and platforms. But given the focus and architecture of generative AI technology today, these truly transformative changes won't happen quickly and few -- if any -- will likely occur within the next 10 years. Over this horizon, AI technology will instead primarily increase the efficiency of existing production processes by automating certain tasks or by making workers who perform these tasks more productive."
Some choice quotes (these may seem like a lot but are a small fraction of the 31-page document):
Daron Acemoglu:
"I began with Eloundou et al.'s comprehensive study that found that the combination of generative AI, other AI technology, and computer vision could transform slightly over 20% of value-added tasks in the production process. But that's a timeless prediction. So, I then looked at another study by Thompson et al. on a subset of these technologies -- computer vision -- which estimates that around a quarter of tasks that this technology can perform could be cost-effectively automated within 10 years. If only 23% of exposed tasks are cost effective to automate within the next ten years, this suggests that only 4.6% of all tasks will be impacted by AI. Combining this figure with the 27% average labor cost savings estimates from Noy and Zhang's and Brynjolfsson et al.'s studies implies that total factor productivity effects within the next decade should be no more than 0.66% -- and an even lower 0.53% when adjusting for the complexity of hard-to-learn tasks. And that figure roughly translates into a 0.9% GDP impact over the decade."
Joseph Briggs:
"We are very sympathetic to Acemoglu's argument that automation of many AI-exposed tasks is not cost effective today, and may not become so even within the next ten years. AI adoption remains very modest outside of the few industries -- including computing and data infrastructure, information services, and motion picture and sound production -- that we estimate will benefit the most, and adoption rates are likely to remain below levels necessary to achieve large aggregate productivity gains for the next few years. This explains why we only raised our US GDP forecast by 0.4pp by the end of our forecast horizon in 2034 (with smaller increases in other countries) when we incorporated an AI boost into our global potential growth forecasts last fall. When stripping out offsetting growth impacts from the partial redirection of capex from other technologies to AI and slower productivity growth in a non-AI counterfactual, this 0.4pp annual figure translates into a 6.1% GDP uplift from AI by 2034 vs. Acemoglu's 0.9% estimate."
"We also disagree with Acemoglu's decision not to incorporate productivity improvements from new tasks and products into his estimates, partly given his questioning of whether AI adoption will lead to labor reallocation and the creation of new tasks."
This section has interesting charts and graphs on which industries have been adopting AI and which haven't.
Jim Covello:
"What $1tn problem will AI solve? Replacing low-wage jobs with tremendously costly technology is basically the polar opposite of the prior technology transitions I've witnessed in my thirty years of closely following the tech industry."
Kash Rangan and Eric Sheridan:
"We have yet to identify AI's 'killer application', akin to the Enterprise Resource Planning (ERP) software that was the killer application of the late 1990s compute cycle, the search and e-commerce applications of the 2000-10 tech cycle that achieved massive scale owing to the rise of x86 Linux open-source databases, or cloud applications, which enabled the building of low-cost compute infrastructure at massive scale during the most recent 2010-20 tech cycle."
"Those who argue that this is a phase of irrational exuberance focus on the large amounts of dollars being spent today relative to two previous large capex cycles -- the late 1990s/early 2000s long-haul capacity infrastructure buildout that enabled the development of Web 1.0, or desktop computing, as well as the 2006-2012 Web 2.0 cycle involving elements of spectrum, 5G networking equipment, and smartphone adoption. But such an apples-to-apples comparison is misleading; the more relevant metric is dollars spent vs. company revenues. Cloud computing companies are currently spending over 30% of their cloud revenues on capex, with the vast majority of incremental dollar growth aimed at AI initiatives. For the overall technology industry, these levels are not materially different than those of prior investment cycles that spurred shifts in enterprise and consumer computing habits. And, unlike during the Web 1.0 cycle, investors now have their antenna up for return on capital. They're demanding visibility on how a dollar of capex spending ties back to increased revenues, and punishing companies who can't draw a dotted line between the two."
Brian Janous:
"Utilities are fielding hundreds of requests for huge amounts of power as everyone chases the AI wave, but only a fraction of that demand will ultimately be realized. AEP, one of the largest US electric utility companies, has reportedly received 80-90 gigawatts (GW) of load requests. Only 15 GW of that is likely real because many of the AI projects that companies are currently envisioning will never actually see the light of day. But 15 GW is still massive given that AEP currently owns/operates around 23 GW of generating capacity in the US. And even if overall grid capacity grows by only 2% annually -- which seems like a reasonable forecast -- utilities would still need to add well in excess of 100 GW of peak capacity to a system that currently handles around 800 GW at peak. The increase in power demand will also likely be hyperlocalized, with Northern Virginia, for example, potentially requiring a doubling of grid capacity over the next decade given the concentration of data centers in the area."
Carly Davenport:
"After stagnating over the last decade, we expect US electricity demand to rise at a 2.4% compound annual growth rate (CAGR) from 2022-2030, with data centers accounting for roughly 90bp of that growth. Indeed, amid AI growth, a broader rise in data demand, and a material slowdown in power efficiency gains, data centers will likely more than double their electricity use by 2030. This implies that the share of total US power demand accounted for by data centers will increase from around 3% currently to 8% by 2030, translating into a 15% CAGR in data center power demand from 2023-2030."
Toshiya Hari, Anmol Makkar, David Balaban:
"AI applications use two types of dynamic random-access memory (DRAM): HBM and DDR SDRAM. HBM is a revolutionary memory technology that stacks multiple DRAM dies -- small blocks of semiconducting material on which integrated circuits are fabricated -- on top of a base logic die, thereby enabling higher levels of performance through more bandwidth when interfacing with a GPU or AI chips more broadly. We expect the HBM market to grow at a ~100% compound annual growth rate (CAGR) over the next few years, from $2.3bn in 2023 to $30.2bn in 2026, as the three incumbent suppliers of DRAM (Samsung, SK Hynix, and Micron) allocate an increasing proportion of their total bit supply to meet the exponential demand growth."
"Despite this ramp-up, HBM demand will likely outstrip supply over this period owing to growing HBM content requirements and major suppliers' supply discipline. We therefore forecast HBM undersupply of 3%/2%/1% in 2024/2025/2026. Indeed, as Nvidia and AMD recently indicated, updated data center GPU product roadmaps suggest that the amount of HBM required per chip will grow on a sustained basis. And lower manufacturing yield rates in HBM than in traditional DRAM given the increased complexity of the stacking process constrains suppliers' ability to increase capacity."
"The other key supply bottleneck is a specific form of advanced packaging known as CoWoS, a 2.5-dimensional wafer-level multi-chip packaging technology that incorporates multiple dies side-by-side on a silicon interposer to achieve better interconnect density and performance for high-performance computing (HPC) applications. This advanced packaging capacity has been in short supply since the emergence of ChatGPT in late 2022."
"We outline four phases of the AI trade. 'Phase 1', which kicked off in early 2023, focuses on Nvidia, the clearest near-term AI beneficiary. 'Phase 2' focuses on AI infrastructure, including semiconductor firms more broadly, cloud providers, data center REITs, hardware and equipment companies, security software stocks, and utilities companies. 'Phase 3' focuses on companies with business models that can easily incorporate AI into their product offerings to boost revenues, primarily software and IT services. 'Phase 4' includes companies with the biggest potential earnings boost from widespread AI adoption and productivity gains."
Summary of key forecasts is on page 25.
