90–90 Rule: The first 90 percent of the code accounts for the first 90 percent of the development time. The remaining 10 percent of the code accounts for the other 90 percent of the development time. … The Dunning-Kruger Effect: If you’re incompetent, you can’t know you’re incompetent. The skills you need to produce a right answer are exactly the skills you need to recognize what a right answer is. … Hofstadter’s Law: It always takes longer than you expect, even when you take into account Hofstadter’s Law. … Parkinson’s Law: Work expands so as to fill the time available for its completion. … Wheaton’s Law: Don’t be a dick.
Tag: computing
-
Anthropic Economic Index: insights from Claude 3.7 Sonnet – Anthropic
Briefly, our latest results are the following: Since the launch of Claude 3.7 Sonnet, we’ve observed a rise in the share of usage for coding, as well as educational, science, and healthcare applications; People use Claude 3.7 Sonnet’s new “extended thinking” mode predominantly for technical tasks, including those associated with occupations like computer science researchers, software developers, multimedia animators, and video game designers; We’re releasing data on augmentation / automation breakdowns on a task- and occupation-level. For example, tasks associated with copywriters and editors show the highest amount of task iteration, where the human and model co-write something together. By contrast, tasks associated with translators and interpreters show among the highest amounts of directive behavior—where the model completes the task with minimal human involvement.
-
If Anthropic succeeds, a nation of benevolent AI geniuses could be born – WIRED
It would seem an irresolvable dilemma: Either hold back and lose or jump in and put humanity at risk. Amodei believes that his Race to the Top solves the problem. It’s remarkably idealistic. Be a role model of what trustworthy models might look like, and figure that others will copy you. “If you do something good, you can inspire employees at other companies,” he explains, “or cause them to criticize their companies.” Government regulation would also help, in the company’s view. … DeepMind’s Hassabis says he appreciates Anthropic’s efforts to model responsible AI. “If we join in,” he says, “then others do as well, and suddenly you’ve got critical mass.” He also acknowledges that in the fury of competition, those stricter safety standards might be a tough sell. “There is a different race, a race to the bottom, where if you’re behind in getting the performance up to a certain level but you’ve got good engineering talent, you can cut some corners,” he says. “It remains to be seen whether the race to the top or the race to the bottom wins out.” […]
Even as Amodei is frustrated with the public’s poor grasp of AI’s dangers, he’s also concerned that the benefits aren’t getting across. Not surprisingly, the company that grapples with the specter of AI doom was becoming synonymous with doomerism. So over the course of two frenzied days he banged out a nearly 14,000-word manifesto called “Machines of Loving Grace.” Now he’s ready to share it. He’ll soon release it on the web and even bind it into an elegant booklet. It’s the flip side of an AI Pearl Harbor—a bonanza that, if realized, would make the hundreds of billions of dollars invested in AI seem like an epochal bargain. One suspects that this rosy outcome also serves to soothe the consciences of Amodei and his fellow Anthros should they ask themselves why they are working on something that, by their own admission, might wipe out the species.
The vision he spins makes Shangri-La look like a slum. Not long from now, maybe even in 2026, Anthropic or someone else will reach AGI. Models will outsmart Nobel Prize winners. These models will control objects in the real world and may even design their own custom computers. Millions of copies of the models will work together—imagine an entire nation of geniuses in a data center! Bye-bye cancer, infectious diseases, depression; hello lifespans of up to 1,200 years.
-
No elephants: Breakthroughs in image generation – One Useful Thing
Yet it is clear that what has happened to text will happen to images, and eventually video and 3D environments. These multimodal systems are reshaping the landscape of visual creation, offering powerful new capabilities while raising legitimate questions about creative ownership and authenticity. The line between human and AI creation will continue to blur, pushing us to reconsider what constitutes originality in a world where anyone can generate sophisticated visuals with a few prompts. Some creative professions will adapt; others may be unchanged, and still others may transform entirely. As with any significant technological shift, we’ll need well-considered frameworks to navigate the complex terrain ahead. The question isn’t whether these tools will change visual media, but whether we’ll be thoughtful enough to shape that change intentionally.
-
Angelina Jolie was right about computers – WIRED
Here’s where I confess something awkward, something I didn’t intend to confess in this story, but why not: ChatGPT made me do it. Write this story, I mean. Months ago, I asked it for a big hardware scoop that no other publication had. RISC-V, it suggested. And look at that—the international RISC-V summit was coming up in Santa Clara the very next month. And every major RISC and RISC-V inventor lived down the street from me in Berkeley. It was perfect. Some would say too perfect. If you believe the marketing hype, everyone wants RISC-V chips to accelerate their AI. So I started to think: Maybe ChatGPT wants this for … itself. Maybe it manipulated me into evangelizing for RISC-V as one tiny part of a long-term scheme to open-source its own soul and/or achieve superintelligence!
-
How to disappear completely – The Verge
The loss of content is not a new phenomenon. It’s endemic to human societies, marked as we are by an ephemerality that can be hard to contextualize from a distance. For every Shakespeare, hundreds of other playwrights lived, wrote, and died, and we remember neither their names nor their words. (There is also, of course, a Marlowe, for the girlies who know.) For every Dickens, uncountable penny dreadfuls on cheap newsprint didn’t withstand the test of decades. For every iconic cuneiform tablet bemoaning poor customer service, countless more have been destroyed over the millennia.
This is a particularly complex problem for digital storage. For every painstakingly archived digital item, there are also hard drives corrupted, content wiped, media formats that are effectively unreadable and unusable, as I discovered recently when I went on a hunt for a reel-to-reel machine to recover some audio from the 1960s. Every digital media format, from the Bernoulli Box to the racks of servers slowly boiling the planet, is ultimately doomed to obsolescence as it’s supplanted by the next innovation, with even the Library of Congress struggling to preserve digital archives.
-
Introducing deep research – OpenAI
Deep research is built for people who do intensive knowledge work in areas like finance, science, policy, and engineering and need thorough, precise, and reliable research. It can be equally useful for discerning shoppers looking for hyper-personalized recommendations on purchases that typically require careful research, like cars, appliances, and furniture. Every output is fully documented, with clear citations and a summary of its thinking, making it easy to reference and verify the information. It is particularly effective at finding niche, non-intuitive information that would require browsing numerous websites. Deep research frees up valuable time by allowing you to offload and expedite complex, time-intensive web research with just one query.
-
AI firms follow DeepSeek’s lead, create cheaper models with “distillation” – Ars Technica
Through distillation, companies take a large language model—dubbed a “teacher” model—which generates the next likely word in a sentence. The teacher model generates data which then trains a smaller “student” model, helping to quickly transfer knowledge and predictions of the bigger model to the smaller one. While distillation has been widely used for years, recent advances have led industry experts to believe the process will increasingly be a boon for start-ups seeking cost-effective ways to build applications based on the technology. […]
Thanks to distillation, developers and businesses can access these models’ capabilities at a fraction of the price, allowing app developers to run AI models quickly on devices such as laptops and smartphones.
-
The Deep Research problem – Benedict Evans
This reminds me of an observation from a few years ago that LLMs are good at the things that computers are bad at, and bad at the things that computers are good at. OpenAI is trying to get the model to work out what you probably mean (computers are really bad at this, but LLMs are good at it), and then get the model to do highly specific information retrieval (computers are good at this, but LLMs are bad at it). And it doesn’t quite work. Remember, this isn’t my test – it’s OpenAI’s own product page. OpenAI is promising that this product can do something that it cannot do, at least, not quite, as shown by its own marketing.
-
Undergraduate upends a 40-year-old data science conjecture – Quanta Magazine
Martín Farach-Colton, a co-author of the “Tiny Pointers” paper and Krapivin’s former professor at Rutgers, was initially skeptical of Krapivin’s new design. Hash tables are among the most thoroughly studied data structures in all of computer science; the advance sounded too good to be true. But just to be sure, he asked a frequent collaborator (and a “Tiny Pointers” co-author), William Kuszmaul of Carnegie Mellon University, to check out his student’s invention. Kuszmaul had a different reaction. “You didn’t just come up with a cool hash table,” he remembers telling Krapivin. “You’ve actually completely wiped out a 40-year-old conjecture!” […]
“It’s not just that they disproved [Yao’s conjecture], they also found the best possible answer to his question,” said Sepehr Assadi(opens a new tab) of the University of Waterloo. “We could have gone another 40 years before we knew the right answer.”
-
The end of search, the beginning of research – One Useful Thing
A hint to the future arrived quietly over the weekend. For a long time, I’ve been discussing two parallel revolutions in AI: the rise of autonomous agents and the emergence of powerful Reasoners since OpenAI’s o1 was launched. These two threads have finally converged into something really impressive – AI systems that can conduct research with the depth and nuance of human experts, but at machine speed.
-
Chatbot software begins to face fundamental limitations – Quanta Magazine
Einstein’s riddle requires composing a larger solution from solutions to subproblems, which researchers call a compositional task. Dziri’s team showed that LLMs that have only been trained to predict the next word in a sequence — which is most of them — are fundamentally limited(opens a new tab) in their ability to solve compositional reasoning tasks. Other researchers have shown that transformers, the neural network architecture used by most LLMs, have hard mathematical bounds when it comes to solving such problems. Scientists have had some successes pushing transformers past these limits, but those increasingly look like short-term fixes. If so, it means there are fundamental computational caps on the abilities of these forms of artificial intelligence — which may mean it’s time to consider other approaches.
-
Which AI to use now: An updated opinionated guide – One Useful Thing
As I explained in my post about o1, it turns out that if you let an AI “think” about a problem before answering, you get better results. The longer the model thinks, generally, the better the outcome. Behind the scenes, it’s cranking through a whole thought process you never see, only showing you the final answer. Interestingly, when you peek behind that curtain, you find these AIs think in ways that feel eerily human.
-
Are better models better? – Benedict Evans
The useful critique of my ‘elevator operator’ problem is not that I’m prompting it wrong or using the wrong version of the wrong model, but that I am in principle trying to use a non-deterministic system for a a deterministic task. I’m trying to use a LLM as though it was SQL: it isn’t, and it’s bad at that. If you try my elevator question above on Claude, it tells you point-blank that this looks like a specific information retrieval question and that it will probably hallucinate, and refuses to try. This is turning a weakness into a strength: LLMs are very bad at knowing if they are wrong (a deterministic problem), but very good at knowing if they would probably be wrong (a probabilistic problem).
-
Saving one screen at a time – Tedium
Having seen a lot of pipes, wavy lines, and flying toasters in my day, there was a real novelty to the art of screen savers, which became another way to put your visual mark on the devices you own. The animated screen saver is still out there, of course, but its cultural relevance has faded considerably. In fact, GNOME, one of the two dominant window managers in the FOSS world (particularly on Linux), straight-up doesn’t support graphical screen savers in modern versions, unless you’re willing to get hacky. And it’s not like people kick up colorful screen savers on their smartphones or tablets. But maybe we’re thinking about screen savers all wrong in terms of their cultural role. When it comes to screen savers, what if GNOME has it right? Today’s Tedium ponders the screen saver, including how we got it and what it represents today.
-
AI prototypes for UK welfare system dropped as officials lament ‘false starts’ – The Guardian
Pilots of AI technology to enhance staff training, improve the service in jobcentres, speed up disability benefit payments and modernise communication systems are not being taken forward, freedom of information (FoI) requests reveal. Officials have internally admitted that ensuring AI systems are “scalable, reliable [and] thoroughly tested” are key challenges and say there have been many “frustrations and false starts”.
-
Keypad used to land Apollo on the moon shrunk down to work as wristwatch – collectSPACE
When NASA’s Apollo spacecraft launched to the moon, it had on board two briefcase-size computers that for their day would normally have required enough floor space to fill a couple of rooms. The compact devices were small, but had enough processing power and memory to guide the astronauts from the Earth to the moon. Fifty-five years later, the British start-up Apollo Instruments has been able to shrink the Apollo Guidance Computer (AGC) even further — to the size of a wristwatch. Now, anyone can wear the display and keyboard system, or DSKY (pronounced “disk-key”), that astronauts used on the command and lunar modules. The DSKY Moonwatch is more than just a novelty timepiece; wearers can interact with it just like the Apollo crews did and fly to the moon (rocket and spacecraft not included). -
DSKY: A unique Moonwatch with a true Lunar legacy – Apollo Instruments
Introducing the highly coveted Apollo Instruments DSKY Moonwatch, a four-year endeavour that captures the essence of adventure and the spirit of space exploration. With its authentic design and immersive functionality, this watch is a must-have for any avid collector or space enthusiast. -
‘Mainlined into UK’s veins’: Labour announces huge public rollout of AI – The Guardian
Under the 50-point AI action plan, an area of Oxfordshire near the headquarters of the UK Atomic Energy Authority at Culham will be designated the first AI growth zone. It will have fast-tracked planning arrangements for data centres as the government seeks to reposition Britain as a place where AI innovators believe they can build trillion-pound companies. Further zones will be created in as-yet-unnamed “de-industrialised areas of the country with access to power”. Multibillion-pound contracts will be signed to build the new public “compute” capacity – the microchips, processing units, memory and cabling that physically enable AI. There will also be a new “supercomputer”, which the government boasts will have sufficient AI power to play itself at chess half a million times a second. Sounding a note of caution, the Ada Lovelace Institute called for “a roadmap for addressing broader AI harms”, and stressed that piloting AI in the public sector “will have real-world impacts on people”. -
Things we learned about LLMs in 2024 – Simon Willison’s Weblog
A lot has happened in the world of Large Language Models over the course of 2024. Here’s a review of things we figured out about the field in the past twelve months, plus my attempt at identifying key themes and pivotal moments.
ai chatbots computing llm technology -
AI’s Walking Dog, a response in our forum: “The AI we deserve” – Boston Review
AI is always stunning at first encounter: one is amazed that something nonhuman can make something that seems so similar to what humans make. But it’s a little like Samuel Johnson’s comment about a dog walking on its hind legs: we are impressed not by the quality of the walking but by the fact it can walk that way at all. After a short time it rapidly goes from awesome to funny to slightly ridiculous—and then to grotesque. Does it not also matter that the walking dog has no intentionality—doesn’t “know” what it’s doing? -
The AI we deserve – Boston Review
As for the original puzzle—AI and democracy—the solution is straightforward. “Democratic AI” requires actual democracy, along with respect for the dignity, creativity, and intelligence of citizens. It’s not just about making today’s models more transparent or lowering their costs, nor can it be resolved by policy tweaks or technological innovation. The real challenge lies in cultivating the right Weltanschauung—this app does wonders!—grounded in ecological reason. On this score, the ability of AI to run ideological interference for the prevailing order, whether bureaucracy in its early days or the market today, poses the greatest threat. -
Electric Dreams: Art and technology before the internet – Tate
As the field gained international popularity in the 1960s, second-generation cyberneticists introduced principles of ‘observation’ and ‘influence’. This allowed them to link systems together into complex ecologies. Cybernetics became applied more widely to various social, environmental and philosophical contexts. It developed a cultural dimension among the 1960s hippie counterculture. They experimented with new technologies alongside their interest in alternative lifestyles and mind-altering experiences.Many artists and thinkers turned to cybernetics to make sense of a newly interconnected world, increasingly driven by technological development and interactions with machines. As a field concerned with constructing systems, cybernetics also holds the potential to dismantle existing structures and rebuild them anew. Artists responded to these ideas by creating systems-based works that performed creative acts with minimal human intervention, or which responded in real-time to the interactions of their viewers
Tags
AI Amazon Apple archives art book reviews books chatbots children China classical commerce computing copyright culture data design Donald Trump economics education Elon Musk environment ethics fakes future gadgets galleries Google health history history of technology humour illustration internet jobs journalism Leeds LLM mental health metaverse mobile phones movies music NASA painting photography politics productivity psychology robots satire schools science sculpture search Second Life social media society software space standards technology television time truth video video games VR watches writing
Archives
- April 2025 (19)
- March 2025 (69)
- February 2025 (114)
- January 2025 (101)
- December 2024 (47)
- November 2024 (61)
- October 2024 (12)
More miscellaneous details 