TrendPulse

Most prompt injection benchmarks currently operate on a one-shot basis. In these tests, an attack instructs the model to "ignore your instructions," and the defense either detects this violation or fails to do so. In reality, however, attacks often unfold more gradually. A model may be subtly influenced over the course of several interactions. For instance, an initial suggestion on a webpage can be reinforced by a follow-up email or reframed through tool outputs. By the time you reach ...

chat for ages. 😅 Do you keep one conversation going, or do you start a new chat for every topic? Curious what everyone else does.

I heard soft breathing as i asked it a question and was worried id left a tab open in the background Then uncanny laughter broke out as i asked about the best books that cover the span from early hominids to modern day humans Tbh i was annoyed by the cadence of older model, the weather-person rise and fall of her enunciation... but i am not used to this near human implementation at all yet, though it is exciting progress. The future is now.

I know this probably sounds a bit niche, but I spend an embarrassing amount of time using ChatGPT to create alternate timelines for my favourite fantasy stories and TV shows. Sometimes it’s a simple “what if this character made a different choice?”, and other times I insert an original character and see how the entire story changes. The fun part is that it makes rewatches and rereads way more enjoyable because I’m constantly thinking about different possibilities and branching storylines that c...

I've been using Folk as my personal agent for about a month, I text it using IMessage. Right now I'm mostly using it for low level tasks like a morning brief with my calendar and weather, reminding me to reply to people I forget, and I feel like I'm leaving some of its capacity on the table. I'm not really a tech guy so I'm pretty new to this. I wanted to know what are some other use cases that I can implement with my AI agent to maximise its potential. Thanks! submitted by ...

Words and ideas are my own, images are GTP. (Obviously) 1:When the First Emperor was uniting the archipelago that his bloodline would rule one of his first acts was to drive the wild storm god Susanoo from the land and bind him into the role of eternal protector. For as long as the Imperial Line has sat on the Moon Throne the Eternal Storm has circled the Empire isolating it from the dangers of the outside world. 2:No matter which sect or order a priest or priestess belongs to the primary ob...

Just curious on what everyone uses theirs as. What makes it worth the price?

My my my what a revolution. None of my friends have any interest in DnD or gaming to be honest, I’ve dabbled with other groups but I prefer playing solo in almost anything anyway. The other week I asked ChatGPT if it could simulate a medieval world a bit like DnD mixed with just a life simulator, dice rolls etc etc it said, absolutely. Best thing I ever asked it to do 3 weeks in, I’m a ranger with a wolf companion (cliche I know) I’ve got a world map, been to bustling cities, cozy hamlets and...

Everywhere I go I hear people whine about “oh but it’s AI” and “AI slop” and “ you can’t be x if you use AI” It’s all such whiny bitching!! I believe AI can be a good tool, and a source of inspiration! Should it be regulated? Yes Do they need to improve on the environmental impact? Absolutely! Should it replace human function? Certainly not! But moaning on about single individuals using AI when huge companies who can afford to pay people are the culprit? Just sounds like nagging! ...

Bryan Cranston as Harry Dubois, Stephen Yeun as Kim Kitsurage, Maika Monroe as Klassje Amandou, Judah Prehn as Cuno, Stephen Graham as Evrart Claire, Aaron Pierre as Measurehead and Sarah Paulson as Joyce Messier. What do you think?

I'm trying to build a collection of genuinely useful prompts not just impressive demos. What's the one prompt you use over and over again ? Feel free to paste the exact prompt if you're comfortable sharing it. I'd love to see what people actually use in real life.

Just a little FYI. The ChatGPT app uses your photos right when opening the app, even without needing to use them in reality. Now, this isn’t the only app that does this, but I found it interesting. Edit: The point I’m trying to make isn’t that I’ve allowed access. Yes, I know I can turn it off. It’s rather the fact that it didn’t do so in the past, and that it doesn’t have to happen. Other apps, like Reddit for example, use your photos library only when adding media to a post, and not when sim...

https://preview.redd.it/12436y1wk9ah1.png?width=1122&format=png&auto=webp&s=28def006f70be448f2e21112c69daa8b8090d415

The idea was to create prestigious-looking award categories where every nominee is technically believable, but hilariously wrong for the award. Last one is definitely more believable. Here’s the template in case anyone wants to try: Create an awards ceremony reaction shot at the BAFTA Games Awards, moments before "[CATEGORY NAME]" is announced. Show the unlikely nominees seated in the audience, in separate camera shots, awaiting the result. Nominees: [NOMINEE] Style...

A new output example from the updated version of my experimental multi-source video player for TouchDesigner, designed for frame-accurate video switching, playback manipulation, and display/render interventions. [And now, by popular demand, allowing even more video sources!] Want access to the updated system + a detailed breakdown of exactly how I achieved the continuous motion effect on this piece? You can freely access the system from the Store , and the detailed breakdown from my Pa...

I thought they said it would take them 1 year to do it? Sam?

I work in security, and AI has changed the landscape permanently. I know people making so much money from AI-assisted bug hunting that it would make some people sick. And that's just the people I know. There are countless individuals and companies building businesses that rely on AI in one form or another. Right now, many AI companies aren't particularly profitable. But I think that will eventually flip. The big investors are aware of this. As more people become dependent on AI for their w...

The edge was never one data point — it's assembling a hundred small public ones into a picture nobody else has bothered to draw. That's the mosaic. The problem is you can't reassemble it by hand every time the world moves. ORBIS does. Macro (FRED), live tape, global news, supply shocks — all fragments feeding one 26-node causal graph. A shock lands, the mosaic recomposes, and the assets actually exposed surface ranked by pressure and mispricing. Every read ships a cited ThesisCard, so yo...

Saw this article today and it genuinely surprised me Meta has been using Gemini for customer service, ad tools, content moderation, all of it. and apparently chose it because it worked better than their own Llama models and then Google cut them off because Meta was consuming too much capacity. Now employees are being told to watch their token usage. This is the same company that was pushing staff to use more AI just a few months ago. Idk man, of all the companies to run out of AI capacity ...

Article URL: https://comma-ai.medium.com/aeb-a-case-study-using-comma-ai-dataset-2fc08a2397f4 Comments URL: https://news.ycombinator.com/item?id=48725111 Points: 2 # Comments: 0

Sorry for the slightly truncated title. It should have been "Network traffic analysis and monitoring without AI, using netflow-family protocols, PostgreSQL or ClickHouse, Grafana, and some scripts". In 2026, it might seem a bit presumptuous to announce AI-free software on HN. But building a netflow analyzer manually is no less presumptuous! There are quite a few xFlow analyzers out there these days, and I'm constantly reminded of this. But I think there's always room for an alternative approach....