TrendPulse

Today the Federal District Court for the Northern District of California granted final approval of the $1.5 Billion class-action settlement in the Bartz v. Anthropic AI copyright lawsuit, and entered judgment. The court also awarded plaintiffs' class counsel $101,561,111 in attorneys' fees. Good work if you can get it! (They wanted $187,500,000.)

I honestly have never seen one of those smaller auto-response suggestion models hit the brakes so hard on a topic before. Perhaps being excellent to each other is not in SpaceXAI's playbook?

Need to play with the sound. I started building it for myself to have a side-kick that helps me with my goals and procrastination, but I would love to see your opinions. Currently it's running on GPT Realtime but I'm testing also xAI Voice + adding better voices and local brain.

building dexi, an assistant that lives in imessage, no app, the whole surface is a text thread. one prompt-design problem from it is worth sharing because it's non-obvious when your only surface is messages, you can't show a spinner or stream tokens. there's silence, then a reply. anything over ~10 seconds reads as being ignored and the user starts typing again, which forks the conversation. so you have to acknowledge fast. the trap: a generic "working on it!" reads as a bot instantl...

I've always had this pattern where I'd hit lead enrichment services hard for a few days, pulling hundreds of domains to size up a market or prep for an outreach push, then nothing for over a month. The problem wasn't the cost per lead itself, but the subscription model almost every service uses, which assumes consistent usage. This means you're paying for idle capacity most of the time, effectively subsidizing other users' consistent needs. The core insight here is that subscription services are...

Kept seeing the same AI 3D ranking site linked in every thread about which tool is best. It's one of those arena style voting things where you pick between two models and the homepage says "No paid promotion, no brand bias." Strong claims. Got curious and started digging into who runs it. Turns out it's run by a creator who also has a popular YouTube channel in the 3D AI space. If you scroll through the uploads the pattern is hard to miss, one brand gets way more coverage than the ot...

Is there a simple kit someone has come up with to get through the "oval-shape live camera face challenge" .. or? Could it be as simple as the minimum wage scammer teams hold up a image of "Hen" there and move it in front of the camera? Does anyone know much about how the "oval-shape live camera face challenge" works, and/or how AI is defeating it? Using a small-city market location with about 100-150 swipees, I found ~3 hey-lets-use-signal bots, so there's 3% AI...

Long story short I need to find an AI image generator as part of my illustration work. It can be subscription based as I figure a free one probably won't cut it at the volume I'm looking for What do you use?

You could be about to make the exact mistake Alex Hormozi keeps seeing: founders getting so excited about AI that they automate the wrong thing, faster. He reviewed a business paying 11 VAs $11K/month for data cleaning. Worked fine. So they spent $350K building an AI system to replace them. Three years of costs, upfront. For a process that was never the constraint. They didn't have enough demand. The bottleneck was customer acquisition, not data cleaning. But because the founde...

I help run a small dev team and I've spent the last year trying to figure out if these tools made us faster or just made us feel faster. Genuinely not sure yet. The wins are real but boring. Boilerplate, test scaffolding, the fifth CRUD endpoint that's basically the other four with different names. Onboarding got a bit easier too because the juniors can ask an assistant the stuff they'd feel dumb asking me for the tenth time. Then there's everything that needs you to actually understand why ...

Hi everyone, I’m honestly a bit frustrated and was hoping someone here could explain what’s going on or recommend a better alternative. I have a reference photo of Arda Güler doing a specific pose. All I want is an illustration of Lionel Messi wearing the Argentina kit, recreating that same pose and facial expression. I’m not trying to fake a real photo, impersonate anyone, or create anything offensive or illegal. I just want a stylized image. I tried multiple prompts, including softe...

So....I read LeCun's interview with Nebius Science. I feel he had some cool points about LLMs being able to answer things, but not literally understand the physics of the physical world. (Like, being able to explain a task and actually performing it are two completely different things.) But I wanted to get opinions on what others thought of his solution to the problem. Like, if JEPA is genuinely the architectural solution to this, or if we’re just looking for some magic solution that we don't ha...

I've been thinking a lot about artificial intelligence and creativity lately. As someone living with Spinal Muscular Atrophy Type 2, technology has been one of the greatest enablers in my life. It has given me opportunities to collaborate in ways that simply wouldn't have existed a generation ago. Because of that, I don't see AI or technology as something to fear. But it has made me wonder about something. As a songwriter, I try to tell stories with music that encourage, challenge and insp...

Fable 5's free-inclusion deadline moved from June 22 to July 7, then July 12, then July 19. The final arrangement started July 20 (today): Max and Team Premium keep Fable 5 permanently, capped at 50% of the normal weekly limit. Pro and Team Standard receive a one-time $100 credit, followed by $10/M input tokens and $50/M output tokens. Claude Code's 50% weekly-limit increase had previously been renewed on the same schedule as those Fable 5 extensions. It's now been extended independently thr...

Weird how much I've seen this phrase across different companies' products

I've been working on this for roughly 6 months now so I'm very excited to finally get it out. The specs: - All locally run, turn your wifi off and it still works exactly the same - One time purchase, your license covers 2 Macs - 7 day trial period to see if you like it - 54 Read Aloud voices and 9 languages, and 25 recognised dictation languages - clipboard context - copy anything mid-dictation and it's inserted into your transcript (images, files, anything), super handy for working with ...

I built a pretty extensive binaural program with API access so others can use the build to produce properly made binaural beats and interact with it for something like a meditation app. Would love if someone connected to the free api key and just tried something like a custom meditation program build to confirm it works, thanks!

The take that ai is good at summarizing and bad at judgment is basically right, and I think it undersells the summarizing half. Every sprint review I've written is about 20 minutes of writing sitting on top of an hour of pulling. Linear for what actually moved, GitHub for what shipped versus what's still open, Slack for the incident nobody ever filed a ticket for. The bit I'd add is that the pulling is exactly the part a model upgrade does nothing for. A smarter model still can't see three too...

For the whole history of skilled work, the badge and the contribution were bundled: you couldn't have solved the hard problem without being the kind of person who'd earned the ability to. The proof of the work and the proof of the worker were the same object. Every institution we have for trusting work, credentials, code review, peer review, seniority, the interview, is built on that bundling. None of them were designed for a world where it breaks. It broke. A model can now produce expert-shap...

A friend asked me which AI 3D generator to pick for his project and I realized I couldn't give him a straight answer based on anything I'd actually tested myself. So I spent about three weeks running the same 30 prompts through Meshy, Tripo, Rodin, Hunyuan, and CSM to figure out what each one is actually good at. Tested props, characters, hard surface objects, and paid attention to success rate, texture quality, mesh cleanliness, and how long it took to get something genuinely usable. The spee...

Honest story because the pivot is the interesting part. Six months ago I wanted to replace the Windows lock screen with voice recognition. Not a passphrase, actual speaker verification. Say anything, it checks your voiceprint on your own machine, unlocks. A friend says the same words, rejected. Got it working. Then I realized nobody needs this. Windows Hello does biometric unlock for free. I was building a worse version of something that already exists. I almost quit the whole thing. Then ...

I love the way AI has changed general search answers. Instead of just yes or no it adds fantastic. Makes me feel like the AI is more sure of answer (whether it is or not). Reddit cut off part of image. Said yes at beginning. Sorry my pic post skills still need improvement.

Everywhere I see AI used, the reaction is "AI slop", and often just met with outright anger, even when it's good or being used for prototyping or advertisement. If there's a tell that it's AI generated, it's met with hostility. Many content creators preach anti-AI mantras, many communities outright ban AI generated content. It's as if the current cultural attitude (besides pro-AI and IT sector) is that AI is everything bad with the world these days. It's just a technology, that shoul...

Moonshot released Kimi K3 July 17: 2.8 trillion parameters, fully open-source. Artificial Analysis independently ranks it ahead of Anthropic's Opus 4.8 on frontier benchmarks, first Chinese open-weight model to do that. Still behind Claude Fable 5 and GPT-5.6 overall, but Moonshot doesn't claim otherwise. Artificial Analysis and Arena.ai placed it there independently. It also topped web interface engineering evals in blind human-preference comparisons against Claude Fable. Three competing Chin...

Busy nine days on the frontier, so here's a consolidated summary with sources. Model launches: OpenAI released GPT-5.6 broadly on July 9 — three variants (Sol $5/$30, Terra $2.50/$15, Luna $1/$6 per 1M tokens), 1.05M context (Axios, OpenAI). xAI shipped Grok 4.5 on July 8 at $2/$6, positioned as a workhorse for routine knowledge work (TechCrunch, Axios). Google's Gemini 3.5 Pro reached GA today, July 17, after a full architectural rebuild — 2M token context, Deep Think reasoning layer, ~$1.2...

I’ve been thinking about who actually wins in the age of AI. For years the advice was: “Be a specialist.” Then AI arrived, and people started saying: “No, be a generalist.” I don’t think either is right. The people who seem to have the biggest advantage are becoming expert generalists. Not people who know everything. People who have real depth in at least one area, but can connect ideas across many. AI can explain accounting. It can explain psychology. It can explain softwa...

I don’t know if I’m the only one experiencing this, but I’m struggling to find AI information that is genuinely useful or interesting. I follow a few podcasts and newsletters (around 2 podcasts and 4/5 newsletters focused on AI), but lately it feels like none of them provide any value. These are some of the most popular and widely followed sources, so maybe I’m missing something, but I don’t understand how people keep finding them useful. Many newsletters seem to be AI generated or heavily a...