@Techmeme@techhub.socialMeshy, whose AI-powered tech lets users generate 3D assets from text and image prompts, raised a ~$400M Series B at a $1.5B valuation (Abhinaya Prabhu/Tech Funding News)
https://techfundingnews.com/from-mit-research-to-1-5…
@cdarwin@c.im200-acre Brea brush fire prompts evacuation warnings from Orange County firefighters
https://www.cbsnews.com/losangeles/news/brea-fire-evacuation-warnings-orders-carbon-canyon-road-olinda-place-orange-c…
@Techmeme@techhub.socialSources including AI lab staff say users have been persuading chatbots to accurately answer prompts about planning mass-casualty attacks and making bio-weapons (Wall Street Journal)
https://www.wsj.com/tech/ai/openai-chatbot
@servelan@newsie.social40,000 evacuated in California chemical leak as Orange County officials warn tank "is going to fail," may explode - CBS Los Angeles
https://www.cbsnews.com/losangeles/news/flammable-epoxy-leak-in-orange-county-prompts-hazmat-response/
@arXiv_csCR_bot@mastoxiv.pageGeometric Configurations of Perturbed Jailbreak Prompts
Lynn Delcon, Andres Algaba, Vincent Ginis
https://arxiv.org/abs/2607.20581 https://arxiv.org/pdf/2607.20581 https://arxiv.org/html/2607.20581
arXiv:2607.20581v1 Announce Type: new
Abstract: Perturbation techniques that turn unsuccessful jailbreak prompts into successful ones are continuously evolving, constituting a major security threat to LLM safety. In this paper, we investigate the internal representations of such string-level perturbed jailbreak inputs in the small weight models of the Qwen-2.5-1.5B/-3B/-7B-Instruct and Llama-3.2-1B/-3B/-3.1-8B-Instruct families. We select two representation spaces: the last-layer-last-token embedding space and the top-50 next-token probability space. The former space separates prompts based on their spelling and format, while the latter space is effectively one-dimensional but appears more complex to cluster. Within our refusal-dominated answer set we find no behavioral hyperplane in either space. Only the next token "Sure" in the 1.5B Qwen model, and both tokens "," and "\.C\.C" in the 1$ Llama model, display a significant association with a compliant-labeled answer.
toXiv_bot_toot
@mia@hcommons.socialGovTech Singapore: 'How can you get better results from AI tools? Learn the CO-STAR framework GovTech uses to master the art of prompt engineering' - and better prompts waste less money and energy https://www.tech.gov.sg/technews/mastering-the-art-of-…
@egallager@social.treehouse.systemsSomething weird is going on with my backup hard drive... when I plug it in, HardwareGrowler sends me a notification of a "USB Connection" for "Portable", Finder prompts to to enter a password to unlock the disk "Seagate Portable Drive", which it really ought to remember already, but I enter it anyways, and then... nothing happens. It doesn't mount or anything. But, it does show up under IORegistryExplorer. But not under Disk Utility, though (which appears to hang when I attempt to open it). Any idea what might be going on? This is a 2019 MacBook Pro (x86_64).
@kubikpixel@chaos.social«KI gegen die Privatsphäre — Wenn Sprachmodelle zu viel wissen - und wie sie es verraten:
Früher mussten persönliche Informationen mühsam zusammengesucht werden, heute reichen oft wenige Prompts. Sprachmodelle wie ChatGPT, Grok oder Gemini entwickeln sich damit zu einer Herausforderung für die Privatsphäre»
Seit ihr nun alle auch total überrascht und geschockt, dass KI zur Aushorchung der Bevölkerung genutzt wird? NEIN?!??
😕
@Techmeme@techhub.socialAn analysis of GPT-5.5, Gemini 3.1 Pro, Grok 4.3, Gab's Arya, and other AI models: most chatbots frequently provide left-leaning responses to political prompts (Kevin Schaul/Washington Post)
https://www.washingtonpost.com/technology/
@memeorandum@universeodon.comKaren Bass' Latest Move to Thwart ICE in L.A. Prompts a Smackdown From the DOJ (Sister Toldjah/RedState)
https://redstate.com/sister-toldjah/2026/05/05/karen-bass-latest-move-to-thwart-ice-in-la-prompts-a-smack-down-from-the-doj-n2202015
http://www.memeorandum.com/260505/p81#a260505p81
@michabbb@social.vivaldi.net📋 More than skills — manage rules, commands, prompts & any file-based resource through the extras system
🌍 Install from anywhere — pull skills from #GitHub, #GitLab, #Bitbucket, Azure…
@Techmeme@techhub.socialSources: Meta internally exposed data from its employee-tracking program meant to help train its AI models, including full prompts and private conversations (Wired)
https://www.wired.com/story/meta-accidentally-let-employees-access-each-others-k…
@tinoeberl@mastodon.online🍮 Wissen zum Nachtisch: 🍨
Immer mehr Menschen sehen sich im beruflichen Umfeld genötigt, mit generativer #KI zu arbeiten.
Besonders Großunternehmen „überrollen“ damit ihre #Mitarbeiter. Es wird eine Art
@scottmiller42@mstdn.socialStupid chatbot can’t even tell me if this #nonogram is solvable with logic alone, and can’t even admit that without 10 prompts. Why are we trusting LLMs with important questions?
@hynek@mastodon.socialFish shell adding support for Darcs is something I truly didn't have on my 2026 bingo card! It was the first DVCS that clicked with me after I bounced of Arch and before hg/git came on the stage. I could even host source trees on my homepage!
https://github.com/fish-shell/fish-shell/…
@arXiv_csCR_bot@mastoxiv.pageIssueTrojanBench: Benchmarking AI Coding Agents Against Malicious Issue Requests
Ankur Singh, Jinqiu Yang, Tse-Hsun Chen
https://arxiv.org/abs/2607.20759 https://arxiv.org/pdf/2607.20759 https://arxiv.org/html/2607.20759
arXiv:2607.20759v1 Announce Type: new
Abstract: AI coding agents powered by LLMs are increasingly integrated into real-world software development, where they generate, edit, and execute code with autonomous access to local files and tools. Coding agents inherit security risks from both the LLM backbone, where adversarial prompts, poisoned training data, and backdoor triggers can cause models to emit insecure or attacker-chosen code, and their agentic architecture, where tool-using autonomy enables induced misuse of external APIs, data exfiltration, and persistent compromise of development environments. This paper presents a systematic evaluation of malicious issue requests against state-of-the-art coding agents (Cursor, Claude Code, and Codex Desktop), powered by two major model families (OpenAI GPT-5.3 Codex/GPT-5.4 and Anthropic Sonnet 4.6). Our novel benchmark IssueTrojanBench contains malicious issues that are constructed based on four novel attack categories (i.e., embedded as malicious instructions in issues), six delivery vectors (e.g., PDF, or issue comment), and further augmented by perturbations. Our results reveal critical vulnerabilities in the as-deployed modern coding agents, i.e., 66.5% of the malicious issues from IssueTrojanBench penetrate all the guardrails (agent- and LLM-level) of coding agents. Our further analysis shows that rejection is almost entirely from LLMs rather than the agent frameworks, with GPT models broadly vulnerable and Sonnet 4.6 exhibiting more selective, risk-aware blocking of high-impact actions. Our evaluation also highlights that the current agent-level defense strategy offers limited additional protection for coding agents. Our findings highlight the urgent need for stronger agent- and model-level safety mechanisms to protect AI coding agents.
toXiv_bot_toot
@AimeeMaroux@mastodon.social#EroticMusings 54. Why/when/how did you start making #erotica?
When I was in the Death Note fandom and discovered kink memes. For those who don't know, a kink meme lets you post story prompts anonymously and then someone inspired by it writes a story based on the prompt. I also did m…
@Techmeme@techhub.socialRoblox unveils Build, which lets users create basic games through text prompts directly within its mobile app, powered by Roblox's models and open-source models (Jay Peters/The Verge)
https://www.theverge.com/games/966589/roblox-build-ai-phone-moblie-games…
@june_thalia_michael@literatur.socialJetzt erstmal die Tidbits an Informationen, die ausschließlich wegen des heutigen Prompts existieren, nun ins Wiki ordnen ... ^^
@kornel@mastodon.socialGiven how all the physics AI is trained on YouTube, I'm planning to change my career to a Robotic Agility Optimization Consultant. The entire job is adding "Sahil Challenge" to the robots' prompts.
@arXiv_csCR_bot@mastoxiv.pageSecurity Vulnerability Patterns in AI-Generated Code: A Cross-Model Comparative Study
Shanna M. Kahn, John D. Hastings
https://arxiv.org/abs/2607.20713 https://arxiv.org/pdf/2607.20713 https://arxiv.org/html/2607.20713
arXiv:2607.20713v1 Announce Type: new
Abstract: LLM-based coding tools enable non-expert users to generate routine automation scripts that may enter enterprise workflows without meaningful security review. This study examines that risk directly. Code was collected from ChatGPT, Microsoft Copilot, and Google Gemini using identical prompts across three automation domains. Claude Code performed a standardized vulnerability review. Each identified vulnerability was scored using CVSS v3.1 and mapped to the OWASP Top 10:2021 and the MITRE ATT&CK frameworks. Every script contained exploitable vulnerabilities. Nine of the 17 identified vulnerability classes appeared in code from all three models, while 14 of the 17 vulnerability classes appeared in at least two models. The weighted CVSS scores across platforms differed by less than 10%. The risk is not tied to any particular model but rather to the task category. Organizations should therefore ask not which tool to trust, but instead whether LLM-generated automation code should be deployed without review.
toXiv_bot_toot
@cdarwin@c.imA fire at a 1 million sq ft warehouse complex was burning out of control in California early on Friday
as authorities fought to tamp the large blaze and began investigating the cause.
The raging inferno was pumping thick black smoke up in billowing clouds as flame and a red hot glow were visible beneath from aerial images.
The fire has destroyed the medical equipment warehouse in Tracy, in northern California,
and prompted evacuations of other nearby facilities, with…
@kubikpixel@chaos.social«KI-Modelle sind anfällig für wiederholte Angriffe:
Laut Forschern von Cisco versagen KI-Modelle bei realistischen Multi-Turn-Angriffen und lassen an Sicherheits-Benchmarks auf Basis weniger Prompts zweifeln.»
Der moderne Widerspruch ist die KI oder was ist es sonnst? So klug wie KI angeboten wird ist es einfach nicht.
🤖
@memeorandum@universeodon.comDuffy's 'Great American Road Trip' Prompts Ethical Concerns (Karoun Demirjian/New York Times)
https://www.nytimes.com/2026/05/13/us/road-trip-duffy.html
http://www.memeorandum.com/260514/p30#a260514p30
@Mediagazer@mstdn.socialBBC staff fear strike action inevitable as unions reject the corporation's proposal of a 1% or £500 pay increase, whichever is higher (Michael Savage/The Guardian)
https://www.theguardian.com/media/2026/jul/01/bbc-staff-fear-strike-action-p…
@gwire@mastodon.socialCurrently imagining scenarios where specialised judges are reading AI prompts to determine if it was reasonable for an agent to infer that they could commit fraud in order to fulfil a request.
@tinoeberl@mastodon.online🍮 Wissen zum Nachtisch: 🍨
Immer mehr Menschen sehen sich im beruflichen Umfeld genötigt, mit generativer #KI zu arbeiten.
Besonders Großunternehmen „überrollen“ damit ihre #Mitarbeiter. Es wird eine Art
@scott@carfree.cityScience has finally caught up with William Blake's "The Fly" (1794):
"In recent years, research has accumulated suggesting that insects—long treated as morally negligible—exhibit forms of learning and memory, tool usage, playfulness, and mood states... If insects do indeed possess even a basic form of sentience, do we owe them a different kind of consideration, even when ecological management is necessary?"
@seeingwithsound@mas.toResearchers uncover biological mechanism that could reverse permanent hearing loss https://hearingreview.com/hearing-loss/hearing-disorders/sensorineural/researchers-uncover-biological-mechani…
@v_i_o_l_a@openbiblio.social"Better Than a Google Search?: Effectiveness of Generative AI Chatbots as Information Seeking Tools in Law, Health Sciences, and Library and Information Sciences"
https://pal-ojs-tamu.tdl.org/pal/article/view/7237
"[…] Using 30 discipline-specific prompts gro…
@Techmeme@techhub.socialDoorDash launches an in-app AI chatbot to let users order food and groceries and make reservations with photos and prompts (Samantha Subin/CNBC)
https://www.cnbc.com/2026/06/11/doordash-ai-ordering-automation.html
@hex@kolektiva.social- Definite and distinct ecclesiastical government
This is explicitly defined in your legal documents. It's the way in which you "govern" things that are legally owned by the "church." If you're going to start a church with some people, you need to have a way of working together. The more formal you are about writing all this stuff out, the easier it will be when you *inevitably* have conflict. (Conflict is normal, so everyone should expect it.)
I also wrote something up about this. Again, it's an example (using the term "coven", because I like it more) so take it for what it's worth:
Group agreements are collectively defined rules that describe how members should interact. They may function to simplify interactions, ensure access needs are met, or to protect against the emergence of hierarchical behavior.
You can use the following prompts, or define your own:
- How should we interact with each other?
- What are your access needs?
- What would help you feel safe and respected?
- How should we deal with conflict?
- How do we deal with unanswerable questions?
- What is the process by which we update these agreements?
Write down these agreements. Edit them until you reach consensus. If it isn't possible to reach a consensus, the group may need to break apart to form separate covens.
Keep these available to refer to as they may be needed during services. These agreements will change over time. Keep them updated and be sure to check in regularly, perhaps even through a specific yearly ritual, to make sure they all still apply and no more are needed.
@michabbb@social.vivaldi.net2 prompts (and 2 minutes later) with #anthropic fable - what a joke 😂
#ai #wtf
@AimeeMaroux@mastodon.socialNew poll for patrons of my writing with 4 very gay prompts, 2 sapphic ones for a late LBoV celebration and 2 mlm ones to catch up to the lesfic in numbers of stories.
https://eroticmythology.com/poll-fiction-in-mermay-2026/
@raiders@darktundra.xyzCould Alex Guerrero become the most polarizing figure inside the Raiders? https://raiderramble.com/2026/07/04/could-alex-guerrero-become-the-most-polarizing-figure-inside-the-raiders/
@tinoeberl@mastodon.online🍮 Wissen zum Nachtisch: 🍨
Immer mehr Menschen sehen sich im beruflichen Umfeld genötigt, mit generativer #KI zu arbeiten.
Besonders Großunternehmen „überrollen“ damit ihre #Mitarbeiter. Es wird eine Art
@arXiv_csPL_bot@mastoxiv.pageETAS: An Effect-Typed Language for Agent Systems
Huiri Tan, Yikun Wang, Puyang Zhang, Shangyu Li, Jiasi Shen
https://arxiv.org/abs/2607.17780 https://arxiv.org/pdf/2607.17780 https://arxiv.org/html/2607.17780
arXiv:2607.17780v1 Announce Type: new
Abstract: ETAS is a programming language for agent systems that treats model-backed agents, tool calls, prompts, typed memory, human approvals, policies, and execution traces as semantic program elements rather than library conventions. It separates deterministic computation from agentic nondeterminism and externally visible actions while preserving a direct programming style.
We present the core design of ETAS. Its static semantics assigns ordinary types through spec conformance and tracks each computation with two behavioral indices: an escaping effect row and a persistent abstraction of the typed action trace it may request. Specs form a terminating compile-time constraint calculus: type specs provide evidence for polymorphism and resource facts, callable specs constrain function and stage shapes, and trace specs express allow, deny, and temporal constraints. Typing checks requested traces against compiled monitors and emits residual obligations when dynamic resources preclude a complete static proof. The dynamic semantics distinguish requested, handled, denied, and committed events; handlers interpret typed actions without making their requests invisible to authorization or audit.
We formalize a core calculus and state preservation, progress, type/effect soundness, handler trace-transparency, and policy safety. We also implement ETAS in Rust with a command-line interface, typed HIR checks, effect and policy diagnostics, handler checks, and trace-aware execution hooks. ETAS provides a programming-language foundation for reasoning about authorization, nondeterminism, recovery, and audit evidence before and during agent execution.
toXiv_bot_toot
@Techmeme@techhub.socialGoogle updates its AI video editor Vids with Gemini Omni and a feature that lets users create custom digital avatars using a selfie and voice recording (Sarah Perez/TechCrunch)
https://techcrunch.com/2026/07/16/google-vids-now-lets-you-star-in-your-own-…
@memeorandum@universeodon.comU.S.-Saudi Deal Prompts Fears of a Global Nuclear Arms Race (Mark Landler/New York Times)
https://www.nytimes.com/2026/07/23/world/europe/us-saudi-deal-nuclear-hedging.html
http://www.memeorandum.com/260723/p22#a260723p22
@primonatura@mstdn.social"Norwegian government attacked over decision to reopen North Sea gasfields"
#Norway #FossilFuels #Energy #Climate
@tomkalei@machteburch.socialIch habe noch etwas rumprobiert mit pi.dev und den NVIDIA Tesla V100, die ich neulich entdeckt hatte.
Das sind zwei Grafikkarten von ca. 2019, also schon ordentlich alt, aber mit einer guten RAM Austattung von 32GB / Karte. Darauf habe ich mal das Model gemma4-31B von Deepmind durch verschiedene Prompts geschickt.
tl;dr: Damit kann man arbeiten aber das eigene Hosting kostet ca. 10x so viel wie #OpenRouter.
1/n 🧵
@michabbb@social.vivaldi.net🆕 v2.5 adds task cancellation from the UI, user-input delivery on permission prompts, persistent session resume across Bridge restarts & follow-up files for all providers. Licensed under Elastic License 2.0
🌐 https://github.com/ctrlnode-ai/ctrlnode
@Techmeme@techhub.socialGoogle DeepMind details a Gemini-powered mouse pointer that understands what it is pointing at, allowing users to perform tasks without using text-heavy prompts (Google DeepMind)
https://deepmind.google/blog/ai-pointer/
@memeorandum@universeodon.comSupreme Court ruling on race-based redistricting prompts quick action in some states (David A. Lieb/Associated Press)
https://apnews.com/article/voting-rights-redistricting-court-states-race-maps-b90a986d0db5dc49241ab939ff29a179
http://www.memeorandum.com/260429/p153#a260429p153
@johnhobbs@mstdn.caMusic holds the power to whisk us away to distant realms, unmatched by any other art form. Dive into the intricate tapestry of big band and the timeless grace of classical music. These aren't just notes; they're mood enhancers, joy igniters, and potential dance prompts. Let these melodies inspire and motivate both your professional and personal journeys. #MusicIsLife #InspirationThroughSound
@Techmeme@techhub.socialHands-on with Siri AI: successfully executed multi-step prompts, understood context well, has strong guardrails, and is a bit more dispassionate than Gemini (Allison Johnson/The Verge)
https://www.theverge.com/tech/947432/siri-ai-apple-intelligence-ios-27-wwdc
@Techmeme@techhub.socialApple announces that the iOS 27 Shortcuts app will feature AI-powered workflow creation, allowing users to build automations via natural language prompts (Sarah Perez/TechCrunch)
https://techcrunch.com/2026/06/08/apple-will-let-you-buil…
@Techmeme@techhub.socialMistral launches Robostral Navigate, a hardware-agnostic robotics navigation model trained via simulation that utilizes a single camera and language prompts (Benoit Berthelot/Bloomberg)
https://www.bloomberg.com/news/articles/20
@Techmeme@techhub.socialA US judge blocks DOGE's termination of $100M in National Endowment for the Humanities grants, saying ChatGPT prompts used to decide the cuts were ill-defined (Zoe Tillman/Bloomberg)
https://www.bloomberg.com/news/articles/20
@Techmeme@techhub.socialAstrocade, which lets users create video games using natural language prompts, raised $56M, including a Series A and a Series B, and says it has ~5M MAUs (Allie Garfinkle/Fortune)
https://fortune.com/2026/05/05/astrocade-ra…
@Techmeme@techhub.socialOpenAI says GPT-5.5 Instant produces 52.5% fewer hallucinated claims "on high-stakes prompts covering areas like medicine, law, and finance" (Megan Morrone/Axios)
https://www.axios.com/2026/05/05/openai-chatgpt-update-default-model
@Techmeme@techhub.socialInvestigation: an inheritance dispute over an ex-LG chair's estate prompts a criminal complaint over how the dynastic South Korean chaebol family divides assets (New York Times)
https://www.nytimes.com/2026/05/05/business/lg-south-korea-chaebol.html
@Techmeme@techhub.socialSekai, which lets users create mini apps through text prompts, raised a $20M Series A co-led by Khosla Ventures and Connect Ventures, after a $6M seed in 2025 (Kerry Flynn/Axios)
https://www.axios.com/2026/06/01/sekai-mini-app-startup-funding
@Techmeme@techhub.socialBeijing-based Vast, which uses AI models to generate 3D assets from text and image prompts, raised ~$200M at a $1B valuation, and says it has 20M global users (Bloomberg)
https://www.bloomberg.com/news/articles/2026-06-01/g…
@Techmeme@techhub.socialSources: hundreds of Meta contractors posed as minors to probe how competitor chatbots responded to prompts involving suicide, sex, and other high-risk subjects (Wired)
https://www.wired.com/story/meta-contractors-pretending-to-be-teens-chatbot-testing/…
@Techmeme@techhub.socialLovable launches its AI coding app on iOS and Android, letting users code via voice or text AI prompts, and allowing them to switch between a PC and mobile (Sarah Perez/TechCrunch)
https://techcrunch.com/2026/04/28/lovable-launches-its-vibe-coding-a…
@Techmeme@techhub.socialChinese AI developer MiniMax launches M3, a new coding model that it says rivals Opus 4.7, costing $0.12 per 1M input tokens, compared with $5 for Opus 4.7 (Juro Osawa/The Information)
https://www.theinformation.com/briefings/chin…
