Guessing Google will add a premium agentic-AI feature to Doppl which when asked "does my butt look big in this?" will quietly cancel your Amazon delivery for that foot-long Toblerone bar you ordered last night. www.theverge.com/news/694087/...

I guess shifting from dropping bunker-busting bombs to dropping F-bombs is a de-escalation of sorts: "Trump swears in angry outburst at both sides over ceasefire violations" www.theguardian.com/world/live/2...

New £14 million project aims to use AI for faster data analysis and to improve medical scans. Canon Medical Research Europe, based in Edinburgh, will showcase AI based smart solutions at Japan expo this week ⬇️ futurescot.com/new-14millio...

Confusing chart makes it look as if 3 of the top 9 councils are on 54%. According to article only 1 is, the next 3 highest are on 52%. It's fine to have 5% "bins" but showing individual columns of datapoints within each bin against the measurement axis misleads. www.theguardian.com/society/2025...

It's remarkable that in the year 2025 the technical gateway that allows or "prevents" (notionally) web scraping of website content is still robots.txt... the digital equivalent of a "keep off the grass" sign, trampled underfoot by at least one US AI model developer. www.bbc.com/news/article...

Drew Purves of Google DeepMind makes a great point that once you commit probabilistic data to a map it will tend to be taken as fact. Like the idea of a confidence interval slider that hides (or marks) data points in the map when they don't meet the essential CI threshhold. youtu.be/vIIIau06wGo?...

Biggest winners from the MS-OpenAI partnership may ultimately be their "divorce" lawyers: "The two companies are in a standoff over OpenAI’s $3 billion acquisition of the AI coding startup Windsurf. OpenAI doesn’t want Microsoft to get Windsurf’s intellectual property" techcrunch.com/2025/06/16/t...

Probably sums up in one sentence why Disney are taking their fight against AI copyright infringement to court: "One search, which could be traced back to a person's Instagram account, asked Meta AI to generate an image of an animated character wearing only underwear." www.bbc.com/news/article...

Quote of the day from Dr. Andres Guadamuz at the University of Sussex, commenting on Disney's legal action against Midjourney for copyright infringement during model training for AI image generation: "It’s Disney, so Midjourney are f---ed, pardon my French." www.newscientist.com/article/2484...

Interesting discussion A.O. podcast. Sarah Isgur & @davidfrenchjag.bsky.social on LLM responses to prompts about US Supreme Court decisions reflecting media coverage of perceived partisan bias rather than reality. Trained on a corpus of 100 books, if 99 say the earth is flat, AI will absorb as fact.

Scotland put ‘at forefront’ of UK’s tech revolution with £750m national supercomputer coming to Edinburgh. Chancellor Rachel Reeves confirms investment in cutting-edge advanced computing facility as part of government spending review ⬇️ futurescot.com/scotland-put...

Charts from the Guardian. Gaps between 1967-1997 and 1997-2027 are both 30 years, yet the year labels aren't evenly spaced. If the trend's the point then why "misplace" 1997? And do the series start 1967 or earlier? In principle simple charts but odd design choices have managed to make them unclear.

Good commitment to human review to ensure any #AI errors are caught: "...it still involves a case worker, who knows the children, going through each individual report to make sense of it to make sure there's not something that's been misunderstood by the AI." www.stokesentinel.co.uk/news/stoke-o...

In "The Accountant 2" there's build up to a bar fight that then happens off screen. But also an extended scene of the main character's brother negotiating, by phone, for a puppy. #BFI are right to worry about AI, but maybe an AI "script doctor" could've brought some consistency to the Accountant 2.

Tour of Britain cycle race in Glasgow today. I don't for sure, but this team with their fancy black and yellow bikes might have the edge!

"The barrister denied using AI in that case but said she may have inadvertently done so in preparation for a separate case where she also cited phantom authorities." #AI users should colour code draft content from LLMs to keep track of what they'll need to check. www.theguardian.com/technology/2...

In the 1980s, during the bitter Iran-Iraq war, Henry Kissinger commented it would be better for the world if both sides lost. Can't think why this came to mind on such a quiet news day in the US.

How to leverage AI agents – your guide to powerful new digital assistants that can transform public services. The Futurescot AI Challenge is a chance to explore the potential of AI agents and Agentic AI for your organisation [partner content with @stormid.com] ⬇️ futurescot.com/how-to-lever...

Google's forthcoming Ask Photos feature sounds like it could be useful when it eventually arrives. Although to the prompt, "find photos of the dog" it's likely to ask if it wouldn't be quicker to just delete the 6 that don't have the dog in them. techcrunch.com/2025/06/04/g...

Scottish AI Alliance launches new practical tools and guidance to support ‘responsible’ AI adoption. Playbook resources designed to help organisations ‘on their journey to implementing trustworthy, ethical, and inclusive AI’ ⬇️ futurescot.com/scottish-ai-...

Edinburgh Napier university partners with security operations centre provider amid rising cyber threat levels. Acumen Cyber's SOC will provide round-the-clock threat detection and monitoring for the university ⬇️ futurescot.com/edinburgh-na...

The discourse around AI and consciousness focusses on the fact if their intelligence is not tied to their physical embodiment in the world, as humans are, then they aren't shaped by true experience, only our digital records. We might soon wish the internet was a nicer place for them to learn from

Great to see progress on these AI initiatives, but hard to over-state the imporrance of improving public sector data quality and interoperability, including between agencies, in order to ensure reliable ouputs and outcomes from use of such tools. www.gov.uk/government/n...

Research suggests that LLMs offering ever expanding context window size for prompt input may provide diminishing returns, relative to effectively implemented RAG for storing content in a vector space for topic-related contextual retrieval. www.datacamp.com/podcast/end-...

Exciting to see pilot councils having the opportunity to improve and refine i.AI's new Minute tool's understanding of local government context and terminology which many of the US-centric out-of-the-box foundation LLMs obviously lack. www.dorsetcouncil.gov.uk/news/just-a-...

Scottish AI firm opens Singapore base as region races to adopt the technology to gain global competitive advantage. Tomoro AI launches Asia Pacific hub to scale up AI among large enterprises in fast-growing sectors including finance and healthcare ⬇️ futurescot.com/scottish-ai-...

Great use of AI/ML. But... in 2023 Nestle UK made a pretax profit of £162M. So I was sceptical of them getting a UK goverment grant of £1.9M to do this. But then... looks like Zest the AI developer is a UK company, so there's logic in encouraging the UK AI sector. www.theguardian.com/environment/...

Focus on innovation for new partnership between NHS Lanarkshire and University of Glasgow. Strategic deal aims to land real-world benefits of research and innovation on healthcare frontline ⬇️ futurescot.com/focus-on-inn...

New guide to help adults develop safe AI systems for kids following consultation with Scottish schools. Learning tool produced by Children’s Parliament to influence how the technology is being built for their generation ⬇️ futurescot.com/new-guide-to...

Does making a persuasive argument in favour of a position come first as a verbal or a written skill? School essay writing teaches us to assemble, evaluate and deploy facts logically. There's a risk if the next generation rely heavily on gen-AI to do this for writing they may weaken key verbal skills

The future of RPA+AI? It's not uncommon for RPA workflows to fail when they encounter edge cases not anticipated in set up. LLMs may provide RPA with more comprehensive scenario generation and pattern analysis from existing historic transaction data. RPA retains advantage for determinstic processing

US newspaper publishes #AI generated article of "summer reads" containing made up book titles it "hallucinated". GPT-4o has been trained on 1.8 trillion parameters but is still lacking just 3 that are crucial for AI to learn to use more often: I. Don't. Know. www.theguardian.com/us-news/2025...

Inside Mission Impossible: Final Reckoning is a good 2-hour movie that's been overstuffed by pointless call-backs and "revelations". Briggs is Jim Phelps' son. Simultaneously implausible and pointless. A pity Chris McQuarrie clings to time-sucking plot strands like Cruise to the bottom of a biplane

Interesting summary by Forbes of the current state of AI coding agents. Confess I got hungup on reference to "workload modernisation" as just "marketing spin frosting". I like the metaphor, but if you spin something, won't the frosting fly off? www.forbes.com/sites/adrian...

NHS Education for Scotland awards £27 million ‘digital front door’ contract to BJSS. Tech company appointed to develop national system for people to access their health and social care records online ⬇️ futurescot.com/nhs-educatio...

Interesting finding by AI team at Bloomberg: given entirely safe RAG context an LLM can break out of its safeguarding (e.g. give advice on commiting crimes). Doesn't mean RAG shouldn't be used, too essential, but model providers must find ways to make safeguards stronger. twimlai.com/podcast/twim...

Sad at the passing of George Wendt, Norm from Cheers. For me Norm and Cliff were the heart of the show. The whole will-they won't-they with Sam (played as a predator?! before Me-Too) and Diane passed me by. But the genuine-seeming chalk and cheese friendship of Norm & Cliff felt far more authentic.

A quote from IBM in 1979: "a computer can never be held accountable, therefore a computer must never make a management decision." Hence all AI vendors emphasise its output must be checked. The risk is checking-the-AI becomes equivalent to "reading the terms and conditions" before clicking "I accept"

Someone told me Waymo's success shows AI has achieved full autonomy. And yet in this podcast their Head of AI says they contunue to struggle with freeways because of the speed and the car can't just randomly stop if it becomes uncertain, like they can do on urban streets. twimlai.com/podcast/twim...

I empathise with the Kidron amendment. I had Dall-E generate 4 images of a dragon on a golden hoard for a slide deck. One looked a lot like Smaug from LOTR. How original would such an image ever be? We all have cultural influences in what we imagine, #AI does too. www.theguardian.com/technology/2...

Interesting divergence between OpenAI's ChatGPT which allows subject-specific custom GPTs to be created and shared, whereas MS Copilot, although its based on the same foundation model, is far less customiseable. MS bets value is in MS365 tenancy/content integration. www.local.gov.uk/case-studies...

Impressive turnaround in performance from only 43% of reports produced within 20 weeks to 83% with use of AI. Includes human-in-the-loop safeguards: "a case worker [who knows the child] went through each individual report to check for any misunderstandings by the AI." www.bbc.com/news/article...

Interesting idea with merit where EBSA is harming pupil attainment. But as with adult WFH, one wonders if the attractions of this remote learning model may see a sudden rise in pupils with EBSA support needs. Waiting lists for the kit, new budget pressures, etc. www.localgov.co.uk/Local-author...

If Trump was right about toy companies being countries then UK's Airfix would defo have the best equipped armed forces. But I'd have to visit Tonka land for the nostalgia. I assumed Tonka was Swedish meaning "child who's lost a finger" but turns out it's a US brand named after a native American lake

Potential tension here between "trained using [data] stripped of personal information" sounds like full anonymisation, versus the individualised interventions which require re-identification. That said, could be real health benefit public happy to opt in for. www.independent.co.uk/news/health/...

Interesting: "The system will not be able to propose relevant case law, which is a high-risk area for LLMs." Hallucination remains AI's Achilles Heel. If any model builder cracks that they'll have a huge advantage over the competition. Looking to reasoning models www.lawgazette.co.uk/news/sra-app...

3 key data points: 1: "Alien", made in UK. 2: "Aliens", made in UK. 3: Adam Sandler's "Jack & Jill", 100% made in California. Don't know about tariffs barriers but there ought to be some sort of cultural exclusion zone thrown around number 3. www.theguardian.com/us-news/2025...

Fusion Fest to headline Glasgow Tech Week at the end of May. The event, in its second year, brings together key tech ecosystem players for the city-region to boost economic growth and spotlight Scotland’s growing reputation as a global innovation hub ⬇️ futurescot.com/fusion-fest-...

Maybe an interesting PhD to be written on why this gender split occurs "a survey of 1,500 jobseekers found that 31 per cent of men were using a paid-for AI tool [in hiring assessments] compared with 18 per cent of women." More emphasis on playing fair, ethical use perhaps? www.ft.com/content/43cd...

OpenAI has had to tone down ChatGPT's agreeableness. Reminded of Douglas Adams and the annoying talking doors on Zaphod's ship, Heart of Gold. Let's hope they don't turn the dial all the way to Marvin-level misanthropy! #UnhappyMedium openai.com/index/sycoph...