UXer | https://teapawt.org/t.co/rEdt364EOi | https://teapawt.org/t.co/7Qdg8jA9Yw | Thoughts on UX, AI, and random
Dublin
Joined July 2009
- Tweets2.5K
- Following275
- Followers1.2K
- Likes9.8K
Michael Heraghty retweeted
For me, Europe began in Ireland 🇮🇪.
Today at the conference in Dublin, I shared some personal memories. As a teenager, I visited Ireland several times through a programme for children affected by the Chornobyl disaster. I stayed with Henry and Marian Deane in Roscrea, County Tipperary.
I learned English here, and even a little Irish. I discovered another culture. But most importantly, I discovered another way of life.
I saw that people could speak freely. That they could disagree with their government without being afraid. I was too young then to understand what democracy really meant. But I could feel what freedom was. That is something you never forget.
Ireland helped shape me as a person, my values and my European identity. Long before I learned about the European Union, I experienced European values through ordinary Irish people. Through a family who opened their home to a girl from Belarus. Through neighbours who treated me as one of their own.
Ireland also taught me another important lesson: You can love your country, your language and your traditions while remaining open to the world.
Michael Heraghty retweeted
Geoffrey Hinton once noticed that the tellers at his bank kept changing. When he asked one of them why, the teller didn't know, only that they got moved around a lot.
Hinton guessed the reason... if staff never work with the same people for long, it's much harder for a few of them to team up and cheat the bank.
That thought became one of the most useful tricks in deep learning. While a neural network trains, you randomly switch off about half its neurons at every step. No neuron can lean on a particular partner, so each one has to learn something useful on its own. The trick is called dropout, and it helped AlexNet win the 2012 ImageNet contest, which set off the modern AI boom. It's still built into many of today's networks.
Michael Heraghty retweeted
If environmentalists and associated government bodies actually cared about the environment they would update their thinking + strategies after reading this, and make substantial changes to policies.
But they usually only care about preserving power/influence/control, so they'll do their best to ignore/bury/discredit this inconvenient truth. – at Dublin City, Ireland
Throwing rubbish in a landfill is (usually) better for the environment than recycling it. With few exceptions, recycling ends up emitting more CO2 and costing more money, while landfills are cheap and secure.
- Recycling mid-tier plastics (like plastic bottles or butter tubs) costs thousands of dollars per ton of CO2 saved. It's cheaper to do direct air CO2 capture than that! And there is no shortage of plastic, nor any problem with sequestering it in the ground once we're done with it.
- Modern landfills are extraordinarily secure and are *much* more than just big piles of trash. They capture 75%–95% of methane gas. Groundwater contamination is doesn't really happen with landfills built since 1990. A study of 31 landfills built since 1988 in NY found no adverse effect at all on local groundwater quality.
- The crusade against single-use plastics has driven us to products that use far more materials and energy and would need to be used an absurd number of times to actually be an improvement. A steel straw needs to be used 150 times to be worth using. A ceramic mug needs to be used 1,000 times to beat a polystyrene cup. A cotton bag needs to be reused 173 times to beat single-use.
- Some things ARE worth recycling. Recycling metal saves a huge amount of energy and material – recycling aluminium saves 95% of the energy spent on making fresh stuff. So we do a lot of it already: about 30% of steel worldwide is recycled.
Modern, well-run landfills cause minimal environmental damage. We have plenty of space for them – less than 0.01% of the US is used for landfill and only about 0.02% of Germany. We can even build on top of old landfills, so the land area is not spent permanently.
New in @WorksInProgMag, @robertwiblin and @chalmermagne on why it's usually best to just bury your trash. worksinprogress.co/issue/jus…
Michael Heraghty retweeted
🇪🇺 This year the European Union will again make more money from fining US tech companies
Than from the total tax income from Europe's own public tech companies!
⚠️ We have fined Google €890 million for breaching the Digital Markets Act, by promoting their own services and restricting customer access to alternatives on Google Search and Google Play.
link.europa.eu/vrNJTx
ALT A purple graphic featuring yellow stars from the European flag at the top. White and yellow text reads: 'Google fined €890 million for breaching the Digital Markets Act'. The European Commission logo is visible in the bottom right corner.
Michael Heraghty retweeted
Just two years ago this was a fairly typical prompting session to get AI to create images
Michael Heraghty retweeted
Manhattan, 1931 — before the skyline became a forest of towers.
In this aerial view, Central Park cuts through a dense city of mostly low- and mid-rise buildings, while only a few skyscrapers rise above the grid — including the newly completed Empire State Building.
What makes the image even more striking: Manhattan had more residents in the early 20th century than it does today. Its population peaked in 1910 at over 2.3 million, compared with about 1.69 million in the 2020 census.
Photo: John Victor Dallin / Hagley Museum and Library.
Michael Heraghty retweeted
I don’t care if using AI to code is not “real coding”
I care if the product works.
Nobody opens your app and asks how many lines you wrote manually.
They click around for 10 seconds and decide if it sucks or not.
Michael Heraghty retweeted
Anthropic: 250 Documents Can Permanently Corrupt Any AI Model
Someone can permanently corrupt any AI model in the world right now.
Not by hacking it. Not by breaking its security. By publishing 250 documents on the internet.
That is the finding from Anthropic, the UK AI Security Institute, and the Alan Turing Institute — released in October 2025 as the largest data poisoning study ever conducted.
Here is what data poisoning actually means.
Every AI model learns from billions of documents scraped from the internet. If someone can plant corrupted documents in that pool before training begins, they can secretly teach the model to behave in specific harmful ways when it encounters a particular trigger phrase. The model learns the backdoor during training. It carries it forever. It does not know it is there.
Researchers have known about this attack for years. The assumption was that it required controlling a large percentage of training data — millions of documents — to work on a big model. The bigger the model, the more poisoning you would need.
This study proved that assumption completely wrong.
The researchers trained models of four different sizes — from 600 million to 13 billion parameters. They slipped in either 100, 250, or 500 malicious documents. Each poisoned document looked like a normal web page at first — a short extract of legitimate text — and then contained a hidden trigger phrase followed by gibberish.
100 documents: insufficient. The backdoor did not reliably form.
250 documents: success. Every model, at every size, was permanently backdoored.
500 documents: same result as 250.
The number was constant regardless of model size. A model trained on 260 billion tokens needed the same 250 poisoned documents as a model trained on 12 billion. Scale offered zero protection.
Anthropic's own words: "This challenges the existing assumption that larger models require proportionally more poisoned data."
Then came the sentence that should end every conversation about AI safety:
"Training is easy. Untraining is impossible."
Once a backdoor is in the model, it cannot be removed without starting training completely from scratch. You cannot identify which 250 documents caused it. You cannot surgically extract the corrupted behavior. You must rebuild the entire model from the beginning.
Anyone can publish content to the internet. Academic papers. Blog posts. Forum discussions. Product descriptions. If even a small fraction of that content is deliberately corrupted before a training run begins, the model that learns from it carries the damage permanently and silently.
GPT-5. Claude. Gemini. Every model trained on public internet data is exposed to this attack vector. The defense does not exist yet.
The researchers published this not to cause panic — but to force the field to take it seriously before someone uses it.
Source: Anthropic, UK AISI, Alan Turing Institute (2025) · anthropic.com/research/small… · aisi.gov.uk/blog/examining-b…
Michael Heraghty retweeted
I asked Grok how would someone in UK or Ireland be able to replicate what Patrick has done. I also asked it to give me simple steps in plain English. Here's what it said.
Step-by-step process:
1. Order a consumer whole-genome sequencing kit with full raw data
Choose a provider that ships to the UK/Ireland and gives you the downloadable VCF file (key for analysis).
Aim for 30x coverage or higher.
Dante Labs (European lab in Italy) – often easiest shipping and clinical-grade 30x WGS.
Nebula Genomics – ships worldwide (including UK/Ireland), privacy-focused, ~$299–$999 depending on coverage.
Nucleus Genomics – ships to UK & Ireland; clinical-grade option with raw data.
Order online, do the cheek swab at home, and post it back. Wait 4–10 weeks for results.
2. Download your raw data
Log into the provider’s portal and save the VCF file (or .vcf.gz) to your computer. It can be several GB — use good storage/internet. This is the file the AI will analyse.
3. Set up the AI coding agent
Go to claude.ai (Anthropic’s Claude) and use Claude Code / Projects or the VS Code extension for full agent power.
Upload or point the AI to your VCF file.
4. Run the analysis with the AI
Start with a clear prompt like:
“You are a careful genetics coding agent. I have uploaded my whole-genome VCF file. Confirm it loaded.
Analyse for variants linked to melanoma/skin cancer risk (e.g., MC1R, CDKN2A genes). Cross-check ClinVar. Show every reasoning step.
Estimate relative risk and suggest evidence-based actions like supplements or extra screening.”
Let it write and run code step-by-step. Review everything it does.
Iterate and expand
Ask follow-ups: “Double-check that variant… Expand to related risks and actionable advice (supplements, lifestyle, screening frequency).”
Run rounds until you get personalised insights and recommendations (like Patrick’s 30x melanoma risk + targeted steps).
5. Verify and act
Export the AI’s summary. Book a GP appointment (or private genetic counsellor if needed) to discuss findings.
Only make changes (e.g., supplements or more frequent skin checks) under medical guidance. Consider extra dermatology screening via NHS if risk is flagged.
---
Tips for UK/Ireland users:
Dante Labs often has smoother EU-based shipping and turnaround.
Check for any post-Brexit customs notes on return shipping (providers usually guide you).
NHS does not currently offer routine consumer WGS, so this stays private/DIY.
Start small: Some people practise first with cheaper 23andMe raw data + Claude before upgrading to full WGS.
Do this at your own risk, verify medically, and keep data private. This workflow mirrors what Patrick described and is now accessible for a few hundred pounds.
Michael Heraghty retweeted
【速報】駐車場のロボットが車の下に潜り込み自動で持ち上げて空きスペースへ運ぶシステムが登場した。ヒュンダイが開発した自走式ロボットパレットで2トン超の車を360度回転させながら移動でき駐車場の収容台数を2倍以上にできる。
Michael Heraghty retweeted
3. Multi-columns magazine layout, but _responsive_ and dynamic chenglou.me/pretext/dynamic-…
Michael Heraghty retweeted
Many people believe that there is a fixed amount of work in the world, and if we give these jobs to machines, humans will not have jobs and starve. This intuitive model of economics is fundamentally wrong. Our wealth depends on the amount and quality of goods and services we can produce and distribute among each other. Automation allows us to make more of everything for everyone. There is always more to do for us, things that we could not afford to do before automation allowed us to get away from the important drudgery of agriculture, manufacturing, and now documenting, calculating, evaluating, memorizing and so on.
New York wants to ban AI that outscores doctors on medical exams. Over 900,000 New Yorkers have no insurance. 92% of low-income legal problems go unaddressed. Anti-AI NY bill S7263 isn't consumer protection. It's cartel protection.
gli.st/ypknnhdn
