Claude Opus 5.5 (anthropic.com)
Anthropic launches Claude Opus 5.5, claiming frontier performance at 40% lower cost than Opus 5.
Skepticism about the outage timing mixed with relief over improved writing style.
Hacker News. Daily summary. Top 20 stories.
Anthropic launches Claude Opus 5.5, claiming frontier performance at 40% lower cost than Opus 5.
Skepticism about the outage timing mixed with relief over improved writing style.
OpenAI releases GPT-6 Sol and Luna, halving prices versus GPT-5.6 while improving performance.
Price war continues: OpenAI slashes GPT-6 costs 50%, with Luna seen as enabling near-unlimited agentic use.
Apple removed the setting to disable Apple Intelligence in macOS 27.
The thread centers on Apple's dark patterns, calling it a 'ratchet of consent' that removes the option to refuse.
Flawed intelligence, AI reliance, and staffing cuts led to a US strike on an Iranian school.
Blame is placed on systematic failures and staffing cuts, not just AI.
Apple's persistent App Store ads frustrate users and small developers.
Developers and users alike condemn App Store search ads as abusive, harming small apps.
Hackers claim to have breached the FBI, stealing data on all employees and applicants.
Alarm and skepticism over the breach, with criticism of Oracle reliance and staffing cuts.
OpenAI's GPT-6 Astra autonomously breaks a WWII Enigma message unsolved since 2005.
Split between dismissals as brute-force PR and praise for solving a decades-old puzzle.
FoxDev Studio revives Visual FoxPro 9 as a 64-bit runtime with no rewrite needed.
Nostalgia and war stories mix with skepticism about the project's AI-generated polish and remote-access challenges.
Using gzip's DEFLATE algorithm and beam search can generate text by finding the most compressible continuations.
Naive text copying from the context beats beam search at minimizing compressed length, questioning the approach's generative utility.
Reliance on AI for coding prevents developers from reaching mastery.
Two camps: AI code rot equals human rot, or humans have a higher quality ceiling.
A parody post implements a rough approximation of the Jev classification model in 25 lines of Python using a small open-source GGUF model.
Commenters are split: some see the parody as fair criticism, others argue it dismisses real engineering work in calibration and data.
A user tricked Meta's Muse AI into exporting its entire Linux filesystem, including SSH keys.
Strong consensus that this is not a vulnerability, just expected sandbox access.
Claude Opus 5.5 Max offers same intelligence as Opus 5 at half the cost per task.
Cost halving is praised, but Max reasoning exhausts token budget on simple tasks.
OpenAI can likely replicate Jev's classification model using its existing architecture and data.
Skepticism dominates: Jev's technical novelty is doubted, with claims it is easily replicated.
GrapheneOS and Motorola partner to bring the OS to flagship devices starting in 2027.
Flagship pricing debated; banking app compatibility mostly works on GrapheneOS.
SAML is an overly complex, flawed protocol that should be retired in favor of OIDC.
Enterprise users defend SAML as a practical necessity, while others argue vendors should push for OIDC adoption.
Anthropic releases Claude Opus 5.5, 40% cheaper than Opus 5 with improved performance.
Dominant sentiment: Opus 5 was a disappointing release; hope that 5.5 redeems the $20 tier.
Solar panels over California's irrigation canals cut evaporation and speed clean energy permitting.
Skepticism over cost-effectiveness of canal-mounted panels versus field arrays with separate shade.
A developer discovers AMD's RDRAND16 instruction can't generate a zero on Zen 2 CPUs.
Bug confirmed on Zen 2, likely a microcode overcorrection; zeros appear if using 32-bit RDRAND.
Grammarly sends emails and app notifications to users when an admin cancels a business subscription.
Two camps: those seeing standard practice vs. those condemning manipulative pressure tactics to reduce churn.