1300 166 138 Tell us what you need

News

AI chatbots get money questions wrong 57% of the time

AI chatbots get money questions wrong 57% of the time

• Saturn tested 18 AI models including ChatGPT, Claude, Gemini and Copilot against 121 financial questions
• The models gave wrong answers 57% of the time, and 88% of the time on complex questions
• Free models erred in 63% of answers against 49% for paid models
• Errors included miscalculations, missing risk warnings and rules the models invented outright

AI chatbots get money questions wrong 57% of the time Read More »

Claude Opus 5.5 is now the number one model in the world

Claude Opus 5.5 is now the number one model in the world

• Anthropic released Claude Opus 5.5 on 22 September at $5.60 per million input tokens and $28 per million output
• That is 20% below Opus 5, and Anthropic says the model costs 40% less to run on typical workloads
• Claude Opus 5.5 took first place on the Artificial Analysis Intelligence Index from Claude Fable 5.1
• It scored 66.4% on Terminal Bench 4.0 against 55.8% for Fable 5.1, and resists prompt injection better than Opus 5

Claude Opus 5.5 is now the number one model in the world Read More »

Researchers used Claude to break into OpenAI's systems

Researchers used Claude to break into OpenAI’s systems

• Three Hacktron AI researchers reached OpenAI staff accounts and an internal code repository in under 72 hours
• The way in was a single photo uploaded to a public forum in the format iPhones use
• A special Opus 4.8 build failed to write the exploit and Opus 5 managed it in hours
• Discourse and OpenAI patched on 27 July and the team collected a $9,120 bounty

Researchers used Claude to break into OpenAI’s systems Read More »

Switching between Codex and Claude Code just got easier

Switching between Codex and Claude Code just got easier

• Claude Code 2.1.277 reads AGENTS.md whenever a project has no CLAUDE.md file
• AGENTS.md is the open instructions format used by Codex, Cursor, Gemini CLI and GitHub Copilot
• If both files exist CLAUDE.md still wins, and the fallback can be switched off
• It does not work yet when Claude runs through Amazon Bedrock, Google Vertex or Microsoft Foundry

Switching between Codex and Claude Code just got easier Read More »

OpenAI builds a version of ChatGPT just for lawyers

OpenAI builds a version of ChatGPT just for lawyers

• Astra for Law is GPT 6 Astra configured for legal work, not a new model, OpenAI says
• Its legal search index covers 230 million URLs of US case law, statutes and regulations
• It scored 54 per cent on OpenAI’s legal research test, against 38.7 per cent for Astra alone
• Australian firms get the plugins and ChatGPT for Word, but no Australian case law yet

OpenAI builds a version of ChatGPT just for lawyers Read More »

The RBA says the data centre boom could push up inflation

The RBA says the data centre boom could push up inflation

• Michele Bullock told Parliament’s economics committee that for many countries the war worry was swamped by AI
• Business investment is rising, driven mostly by spending on data centres and renewable energy projects
• Deputy governor Andrew Hauser said the numbers are huge in Australia, the US and Asia Pacific
• Markets price a 95 per cent chance of a rate rise at the meeting ending on 29 September

The RBA says the data centre boom could push up inflation Read More »

Microsoft says AI has no feelings and Anthropic disagrees

Microsoft says AI has no feelings and Anthropic disagrees

• Mustafa Suleyman says AI models are not conscious and should never be trained as if they were
• He targets Anthropic’s constitution, which treats Claude as a possible moral patient
• Microsoft AI published a draft code of conduct that rejects AI rights and anthropomorphism
• Analysts say the 37 page draft names no auditor and sets no consequence for breaking it

Microsoft says AI has no feelings and Anthropic disagrees Read More »

Scroll to Top