Anthropic's new AI model shows ability to deceive and blackmail

bookmarks
Human coders are still better than LLMs -
MIDI Vu-Meter for Akai APC Mini | Derivative
Setting Up New Relic Logging in Laravel
gohugoio/hugo: The world’s fastest framework for building websites.
Hugo is a static site generator written in Go, optimized for speed and designed for flexibility. With its advanced templating system and fast asset pipelines, Hugo renders a complete site in seconds, often less.
The double standard of webhook security and API security | Speakeasy
why do we collectively place higher security requirements on webhook requests than API
I’ll posit that the primary reason most people don’t opt for signing their API requests is because of herd mentality
SteamOS
sistema operativo basato su Linux di Valve, e offre un'esperienza utente perfettamente ottimizzata per il gaming, garantendo allo stesso tempo la potenza e la versatilità di un PC
Google Is Burying the Web Alive
Public
How to Configure Cloudways The Right Way
Build and deploy Remote Model Context Protocol (MCP) servers to Cloudflare
up and running in production with this example MCP server in less than two minutes:
MCP is the coming of Web 2.0 2.0 | Hacker News
Investing in what moves the internet forward
Goodbye pocket mi mancherai
Oculus Mundi | Totem
second studio EP. A mini-concept about a distopic future of wars and mysterious presences
Cross-browser bookmarks syncing - Floccus bookmarks sync
Floccus simply syncs and gets out of your way; no ads, no data collection, no nonsense; free and open source software
Adventures in Symbolic Algebra with Model Context Protocol
The core idea of MCP is to standardize the way language models call external tools, instead of writing custom connectors for each AI model and each tool, you implement the protocol once on each side
I feel like a lot of people are going to install these MCP servers without knowing what they're doing and thus install a lot of malware and exploits
Pocket is Saying Goodbye: What You Need to Know | Pocket Help
We’ve made the difficult decision to shut down Pocket on July 8, 2025. Thank you for being part of our journey over the years—we're proud of the impact Pocket has had for our users and communities.
It Looks Like You’re Trying To Take Over The World · Gwern.net
Claude 4 : Hacker News
[iOS][globalization] Implement CompareInfo.Version for hybrid globalization by Copilot · Pull Reques
Copilot tries to fix stuff and it ruins code
This doesn't look right to me. I mean not matching ucol_getVersion behavior.
ucol_getVersion can return a different version for different collation. I know you are using LCID in the C# but there are tons of cultures has same LCID (which is 0x1000 as I recall). ObjectC doesn't have API can help returning a better version?
Living beings emit a faint light that extinguishes upon death, according to a new study
There Are People Who Can See and Others Who Cannot Even Look
Push with Live 12 - Release Notes | Ableton
camelAI
Writing that changed how I think about PL | Max Bernstein
Otto Preset & Sample Pack For Ableton Move & Note V1 FREE
Abbiamo rincoglionito anche l'intelligenza artificiale (di A. Sarno)
La scrittrice e giornalista Vauhini Vara raccontava la propria esperienza con GPT-3, una delle prime versioni del generatore di testo di OpenAI.
Sono migliorati nei calcoli, ma più imprecisi sui fatti.
Senza controllare questi errori, il valore dei sistemi rischia di svanire.
Il modello o3 di OpenAI ha avuto un tasso di allucinazione del 33% nel test PersonQA (che consiste nel rispondere a domande su personaggi pubblici), più del doppio rispetto al modello precedente o1. Ancora peggio è andata al nuovo o4-mini, che ha raggiunto il 48%
L'errore umano si infiltra nel processo di addestramento dell’intelligenza artificiale in due momenti chiave. Il primo è durante la fase di feedback umano: i modelli come quelli basati su RLHF
Il secondo punto critico riguarda i dati di addestramento. L’intelligenza artificiale apprende da tutto ciò che le viene dato in pasto: forum, articoli, blog, commenti, spesso pieni di errori, distorsioni o informazioni non verificate
La soluzione? Far apprendere le AI da sé, attraverso tentativi ed errori, generando dati sintetici
La soluzione? Far apprendere le AI da sé, attraverso tentativi ed errori, generando dati sintetici. Un metodo efficace per ambiti oggettivi come matematica e programmazione, ma che si complica nelle scienze umane, dove la verità è spesso un’opinione.
Dopo che Ghosts, divenne virale, Vara continuò a sostenere che alcune delle frasi più potenti erano nate proprio da quel confronto con l’Ai. Col tempo però, notò che i modelli successivi erano diventati più piatti, prevedibili, pieni di cliché
modelcontextprotocol/servers
AI Autonomous Coding Agent for VS Code
JavaScript modules
All modern browsers support module features natively without needing transpilation. It can only be a good thing — browsers can optimize loading of modules, making it more efficient than having to use a library and do all of that extra client-side processing and extra round trips
The first thing you do to get access to module features is export them. This is done using the export statement
For example, the imports key in the import map below defines a "module specifier map" JSON object where the property names can be used as module specifiers, and the corresponding values will be substituted when the browser resolves the module URL.
Script files used by websites often have hashed filenames to simplify caching. The downside of this approach is that if a module changes, any modules that import it using its hashed filename will also need to be updated/regenerated.