标签：#ai

Rockstar Games will allow players to upgrade older versions of Grand Theft Auto V to PlayStation 5 and Xbox Series X / S for free just months before the launch of GTA VI. Starting June 18th, players with any GTA V copy on PS4 or the digital version on Xbox One can get the current-gen […]

"How does ChatGPT think ?" It doesn't. The entire mechanism behind every chatbot is almost anticlimactic: it predicts one next word , adds it, and repeats. I built a tiny interactive predictor so you can be the model — and it explains both the magic and the flaws. 🔮 Be the model: https://dev48v.infy.uk/ai/days/day6-next-token.html This is Day 6 of AIFromZero — AI literacy, one concept a day, no code to follow. 1. It only predicts the NEXT word Given everything so far, the model outputs a probability for every possible next word, picks one, appends it, and runs again with the longer text. Paragraphs, code, poems — all of it is this one step on repeat. "the cat sat on the ___" → P(mat) high, P(bird) low 2. It's a probability over the WHOLE vocabulary The output isn't one word — it's a number for every word it knows (100,000+ for a real model). Most are near zero; a handful are plausible. The bars in the demo are that distribution, over a tiny vocabulary. 3. Autoregression: feed the output back in After picking a word, it becomes part of the input for the next prediction. Predict → append → predict again. Because each new word conditions on all the previous ones, short local choices add up to coherent long text. 4. Temperature = the creativity dial Once you have probabilities, how do you choose? Temperature reshapes them before sampling: Near 0: the top word always wins — safe, repetitive. High: the odds flatten, so rarer words get a real chance — creative, error-prone. p = p ** ( 1 / temperature ); // then renormalise and sample Drag the slider in the demo and watch the bars sharpen or even out. That one knob is what an API calls "creativity." 5. Where do the probabilities come from? In my toy, from counting which word followed which in a few sentences (a "bigram" with 1-word memory). A real LLM replaces the counting with a giant neural network trained on much of the internet, and its memory spans thousands of words. The mechanism is identical — only the quality of th

Pick the wrong loss function and your model optimises the wrong thing — perfectly. The loss is the single number training tries to shrink, so it quietly defines what "wrong" even means. I built an interactive visualiser of MSE, MAE, and cross-entropy so you can see why the choice matters. 🎯 Drag the prediction: https://dev48v.infy.uk/dl/day6-loss-functions.html This is Day 6 of DeepLearningFromZero. Loss = one number for "how wrong" The network's output is compared to the truth and collapsed into one scalar. Everything in training exists to make that number smaller. Choose the loss and you've defined the network's entire goal. MSE — square the error (regression) const mse = ( pred , y ) => ( pred - y ) ** 2 ; Squaring means off-by-4 hurts 16×, off-by-1 hurts 1×. MSE obsesses over large errors — great when big misses are unacceptable, risky when outliers will drag the model around. MAE — absolute error, outlier-robust const mae = ( pred , y ) => Math . abs ( pred - y ); Linear penalty: off-by-4 hurts exactly 4× off-by-1. One wild outlier can't dominate. The trade-off is a constant gradient, so it can be slower and less precise near the answer. Cross-entropy — for classification When the output is a probability, you don't use MSE. Cross-entropy rewards confident-and-right and brutally punishes confident-and-wrong: const bce = ( p , y ) => - ( y * Math . log ( p ) + ( 1 - y ) * Math . log ( 1 - p )); Predict 1% for the true class and the loss screams toward infinity. In the demo, switch to Classification and slide p toward 0 to watch it explode. The slope is what learning actually uses Backprop doesn't follow the loss value — it follows the loss's gradient (slope) downhill. That's why the shape matters: cross-entropy's steep slope when very wrong gives a strong corrective push, helping classifiers learn faster than MSE would. grad = dLoss / dPred ; // gradient descent steps along this Choosing the loss is a design decision Predicting a price? MSE or MAE. Yes/no? Binary

Naive Bayes ran real spam filters for years, and it's the rare ML model whose "training" is just counting . No gradient descent, no iterations — count words, apply Bayes' rule, multiply. I built one from scratch and visualised exactly which words push a message toward spam. 📨 Interactive demo (type a message): https://dev48v.infy.uk/ml/day6-naive-bayes.html This is Day 6 of MachineLearningFromZero — algorithms from scratch, no scikit-learn. 1. Bag of words — order doesn't matter Naive Bayes treats a message as a set of words. "free cash now" and "now cash free" look identical to it. That throws away grammar, but for spam detection the words present matter far more than their order — and it makes the math tiny. 2. Training = counting For every word, how often does it appear in spam vs ham? for ( const { text , label } of trainingData ) for ( const w of tokenize ( text )) counts [ label ][ w ] = ( counts [ label ][ w ] || 0 ) + 1 ; free and click flood spam; meeting and tomorrow live in ham. One pass over the data, done. 3. Bayes' rule flips the question You measured P(words | spam) , but you want P(spam | words) . Bayes flips it: P(spam | words) ∝ P(spam) × P(words | spam) P(spam) is the prior (how common spam is); the likelihood multiplies in the word evidence. 4. "Naive" = pretend words are independent The trick that makes it fast: assume each word is independent given the class, so the likelihood is just a product: P(words | spam) = P(w1|spam) × P(w2|spam) × ... Real words aren't independent ("credit" and "card" co-occur), so it's a naive lie — but the classification still lands right astonishingly often. 5. Smoothing + logs keep it stable Two practical fixes. Add 1 to every count (Laplace smoothing) so an unseen word doesn't zero out the whole product. And add logarithms instead of multiplying tiny probabilities, which would underflow to 0: score [ label ] = Math . log ( prior [ label ]); for ( const w of words ) score [ label ] += Math . log (( counts [ label ][

“People just really want to take back control of their time, their lives, their attention... They’re down for whatever helps them do that.”

If physical AI is going to match the accomplishments of LLMs, there's a data problem that needs to be solved.

Lego's newest set is a fully functional miniature pinball machine.

Horror is having a moment. In 2026, the genre is especially well-represented: new blood is dominating the box office through films like Backrooms and Obsession, established names like Sam Raimi and Damian McCarthy are at the top of their game, and long-running franchises like 28 Years Later and Resident Evil continue to stay relevant. But […]

Pramaana will focus on highly sensitive verticals like law, drug discovery, and tax preparation — where errors can be costly and reliability is at a premium.

The Toy Story franchise began with a story about a vintage doll feeling threatened by the arrival of an electronic action figure. Woody and Buzz's rivalry embodied a shift that was happening in the '90s as children's toys were becoming more technologically sophisticated, and while toys have gotten even more tech-focused in the years since, […]

Critical Energy is turning rocket engines into geothermal power plants, and it wants to build 300 GW per year by 2045.

AI is pretty reliable at putting things on your calendar these days, but it hasn't quite cracked answering the related and all-important question of "What should I do this weekend?" Meta's new AI Mode in search could be a useful tool - if it ever learns to stop getting stuff wrong. AI Mode is a […]

Turbo Start Aisle is our agentic e-commerce starter on Shopify and Sanity. Instead of filters...

If you build with AI, three stories this week rhyme into one theme: the hype is colliding with the bill. Here's the builder's read on each — and what I'd actually do about it. 1. Most of a new TikTok feed is now AI slop A Kapwing study reported by Tubefilter hand-checked 10,742 videos across 20 categories and found that 59% of what a brand-new TikTok account sees is AI-generated . Kids content was the worst — 57% slop, with the #CartoonKids tag hitting 97% — and TikTok serves roughly 3x more slop than YouTube. Why builders should care: generation is now free and infinite, so volume is worthless as a moat. The scarce thing is taste and verification. If your product or content can be faked by a feed of bots, it will be. Polish, point of view, and "a human clearly did this" are the new differentiators. 2. Databricks grew 80% — but agents are eating its margins Per CNBC , Databricks' annualized revenue jumped about 80% to ~$6.9B, and its AI products now bring in $1.7B (up from $1.4B). The catch: the CEO says gross margin "will go lower" as customers run more agents. Why builders should care: this is the quiet tax of agentic software. An agent that loops, retries, and calls tools burns far more tokens than a single API call. If you're shipping agents, budget for inference at scale , not the sticker price on the pricing page. Profitability now lives in prompt efficiency, caching, and knowing when not to call the model. 3. 60% of US consumers are turned off by "AI" branding A WordPress VIP survey of 2,000 people, covered by TechCrunch , found that 60% reject "AI" in brand messaging , while 86% still want to check the original sources behind a claim. Why builders should care: "Now with AI!" is starting to read like a warning label. Sell the outcome, not the technology — "2x faster," "fewer errors," "your data stays private" — and cite where your results come from. Trust is becoming a feature you ship, not a slogan you bolt on. The takeaway Feeds are flooded, agents are cost

Lars Faye's Agentic Coding Is a Trap — published Sunday, May 3, picked up on Hacker News at 398 points and 316 comments — is the best single compendium of the cognitive-debt evidence base anyone has put together in 2026. It catalogues the studies. It names the trade-offs. It lands on a personal-discipline conclusion. The receipts are now collected; the careful reader will have spent the weekend nodding through them. Buried in Faye's second paragraph, almost in passing, is the line that does the actual analytical work. Faye describes the agentic workflow as a process in which "someone defines the project's requirements ... generates a plan, and then pulls the slot machine lever over and over, iterating and reiterating with often multiple agent instances until it's done." The link goes to a March post by Quentin Rousseau, CTO and co-founder of Rootly, titled One More Prompt: The Dopamine Trap of Agentic Coding. The metaphor isn't Faye's. Rousseau got there first, in clinical language: the workflow runs on "variable ratio reinforcement — the same psychological mechanism that makes slot machines the most addictive form of gambling" . That is the framing the rest of Faye's piece is downstream of, and it is the framing this article is about. What the receipts add up to Faye's catalogue, briefly. Anthropic's own research note on internal use names what it calls the "paradox of supervision" : effective use of Claude requires the very skills that sustained Claude use atrophies. MIT Media Lab's Your Brain on ChatGPT measured the cognitive impact and labelled it cognitive debt . A Microsoft study covered by 404 Media reached parallel findings for knowledge workers more broadly. A separate Anthropic study on coding skills reported a 47% drop-off in debugging skills among engineers leaning heavily on AI-assisted workflows. Sandor Nyako, the LinkedIn engineering director who oversees fifty engineers, has reportedly asked his team not to use these tools for "tasks that require cri

Google's first new smart speaker in six years starts shipping on June 29th, narrowly missing its promised spring launch window. Preorders for the Google Home Speaker open today, June 17th. Nothing has changed hardware-wise in the nine months since the $99 speaker was announced. It has the same slightly squished round design, with touch-capacitive buttons […]

OpenTelemetry has become the standard for observing modern systems. But when you start building AI applications, traditional traces aren't enough. You don't just want to know that a request happened. You want to know: Which model generated the output? Which provider was used? How many tokens were consumed? What embedding model processed the documents? How much did the operation cost? These questions become even more important when building Retrieval-Augmented Generation (RAG) systems. Recently while contributing to Mastra, I discovered an observability gap involving RAG embedding operations. This led me to open a pull request that introduced proper OpenTelemetry GenAI semantic mappings for RAG_EMBEDDING spans. The Problem Mastra already exported rich metadata for several AI operations. However, RAG embedding spans were missing standardized GenAI semantic attributes. As a result, observability tools could see that an embedding operation occurred, but they couldn't easily understand: Model information Provider information Token usage Embedding-specific metadata Without standardized semantic conventions, dashboards and tracing systems lose valuable context. This becomes a bigger issue in production environments where teams need visibility into AI workloads. Understanding RAG Embedding Spans A typical RAG pipeline looks like this: Documents ↓ Chunking ↓ Embedding Model ↓ Vector Database ↓ Similarity Search ↓ LLM Generation The embedding stage is critical. Every document chunk gets transformed into a vector representation. If observability data from this stage is incomplete, debugging performance issues becomes significantly harder. Why OpenTelemetry Semantic Conventions Matter OpenTelemetry doesn't just define traces. It also defines semantic conventions. These conventions create a common language for telemetry data. Instead of every framework inventing custom field names, everyone follows the same standard. For GenAI workloads this means tools can automatically underst

I Built a Vanilla JS Web App that Scored 100/100 Under Lighthouse’s New "Agentic Browsing" Audit. Here’s What It Means. If you have run a performance audit on PageSpeed Insights or Lighthouse recently, you might have noticed a fascinating new line item quietly slipping into the metadata report: Agentic Browsing . When I audited my free tool suite, Paktheta , I managed to hit the ultimate developer milestone— a perfect 100/100 across Performance, Accessibility, Best Practices, and SEO. But seeing that perfect score alongside the label "Agentic Browsing" got me thinking. What exactly is an AI-driven agent experiencing when it hits our sites, and why is this the new gold standard for web performance? Let's dive into what Agentic Browsing actually means for the future of optimization. What on Earth is "Agentic Browsing"? Historically, speed tests like Lighthouse were passive. A headless browser opened your URL, waited for the page to load, recorded metrics like First Contentful Paint (FCP) and Largest Contentful Paint (LCP), and closed the tab. It was a linear, predictable, and frankly synthetic snapshot. Agentic Browsing changes the paradigm entirely. Instead of a basic static script, modern auditing platforms use autonomous, intelligent browser agents. Guided by modern AI-driven browser control (using updated instances like HeadlessChromium), these agents don't just stare at your page—they explore it like a real human would. An agentic audit runner will: Identify interactive buttons and click them to test responsiveness. Scan form elements to see if they accept paste commands cleanly. Intelligently look for broken layout shifts (CLS) by dynamically scrolling and triggering micro-animations. Interact with JavaScript components to see if they block the main execution thread. In short: It simulates real, unpredictable human behavior at lightning speed. If your site relies on bloated frameworks that look fast initially but lock up the second a user tries to interact, an a

The Canadian pension giant will acquire an 8.2% stake in CtrlS, a tech giant that operates more than 15 data centers across India.

With this acquisition, DeepL is opening an office in San Francisco to expand its U.S. business.