Ian Kellar

Reasoning models don't always say what they think Research from Anthropic on the faithfulness of AI models' Chain-of-Thought

We want to believe we know what's going on when we prompt - see the cheat sheets that suggest we can prompt to adjust parameters (no, mentioning temperature is not the same as adjusting the temperature)

www.anthropic.com/research/rea...

Started module getting them to make custom GPTs to produce Cochrane clinical answers. Finished by making an authoring tool, developing a CoT prompt to get a customGPT to guide themselves thru PRISMA2020, and check they had requisite information for each item. #AI #evidencesynthesis #highereducation

Finished teaching. In teams of 8-10, students learn question frameworks, reusing existing search elements, SRAccelerator / Rayyan, summarisation and data extraction using LRM+RAG, MMAT2018, SWIM and PRISMA. Between 2weekSR tools and priority screening, you CAN teach students to do systematic reviews

London, Cambridge, Warwick, Leeds, Exeter, Newcastle, Birmingham, virtual, and Bristol. I can't remember the order even other than London and Cambridge were the first 2.

This week I presented some of my PhD research at the #UKSBM2025 Annual Scientific Meeting. Fantastic experience and great to hear about so much exciting research!

YouTube video by SysteMatic Juggling multiple health conditions: Roberts story

Very powerful video shared by @francesmair.bsky.social in her #UKSBM2025 plenary on the challenges of managing multiple health conditions, particularly on the poverty line.
We have to tackle poverty&inequality to improve our public health.
youtu.be/uJiVUPFFXBw?...

20 years of UKSBM (that's the UK Society for Behavioural Medicine)! 🥳 I was lucky enough to be at a fair few of these. Here's to the next 20 #UKSBM2025

YouTube video by NIHR Applied Research Collaboration South London Social Care ImpRes

Social care and implementation researchers at ARC South London have produced a new guide to designing and conducting implementation research within social care. #ImpSci

👉 arc-sl.nihr.ac.uk/news-insight...

Hopefully UKSBM can move their society accounts over here in the near future. I don't really want to 'join the conversation' on Meta and X thanks... #UKSBM2025

Great call to action in community pharmacy research from Ade Williams @adewilliams.bsky.social #UKSBM2025

UK Universities Funding

"The 'solution' used to be growth – reckless growth. Now it seems it is downsizing – reckless downsizing. In both cases, this is for the business model’s sake, and for the increasing number of business partners — not for students...and not for staff."

#UKHE #AcademicSky

Nigeria currency crisis leaves Dundee University fighting to survive Profuse spending left the institution ill-prepared for a sharp drop in foreign students after Nigeria, a fertile recruiting ground, devalued its currency

Despite drawbacks, several good points here: (1) Dundee University is genuinely fighting for its life; (2) some uni management teams in some unis are out of control; (3) the multiple UK govts don't even have a clue what's going on, no controls and no plan... (1/2) www.thetimes.com/uk/scotland/...

The author misunderstands how LLMs can be useful in writing articles. Using LLMs without using RAG is just relying on the patterns learned in the training of the model. If you know the relevant literature, using RAG, it fetches the relevant info from the papers you give it, adds it to the prompt and generates a response based on the relevant retrieved information. This addresses issues with hallucinations, cut off dates and citations. Tools like ChatGPTs Deep Research, and The-Literature add search to the beginning of this process. My students know that if want an essay out of an LLM, upload the PDFs of the papers in the reading list, the assessment guidance and the marking criteria. If they are using notebookLM, they could listen to a chatty podcast, or chat to the podcast in interactive mode to give them ideas for prompts.

If you can’t train your students to do better than LLMs plus RAG, then their graduate jobs aren’t going to be around for long. Pretending LLMs fed with the relevant literature and trained in a academic writing style (like notebookLM is trained to generate output in a podcast style) is to bury your head in the sand. Knowledge work is going to change and we have to prepare ourselves and our students for what’s coming. If Google trained a LLM on all the lectures that were captured over the pandemic, they could have create an automatic lecture generator, that fed with relevant papers could generate a lecture. It’s so obvious someone will. Like Uber drivers are really training the driving model that will replace them, all our open papers and lectures, not to mention all our feedback on Turnitin, are training the models that will replace us if we can’t to do more than just regurgitate knowledge. This article doesn’t illuminate the work we’ve got ahead of us

This article misunderstands how you can effectively use LLMs for academic writing. The error is common among my colleagues. I teach my students to write using LLM+RAG and writing style prompts. scholarlykitchen.sspnet.org/2025/03/11/g.... Comment is in the image

Will the use of generative AI shift higher education from a knowledge-first system to a skills-first system? - HEPI search.app/DvsRQJAR5r58...

How to trash a University's reputation And demotivate the people you need to help you succeed Leighton Andrews Hey, please come to the ‘reshaped and reduced Cardiff Business School’! It’s not the most obvious marketing slogan, is it? Nor is it a motivating mantra for our academic staff, whatev...

How to trash a University's reputation And demotivate the people you need to help you succeed nation.cymru/opinion/how-...

This is why @kityates.bsky.social post is so relevant today

kityates.substack.com/p/science-is...

and why my dystopian piece is perfectly plausible.

christinapagel.substack.com/p/a-dystopia...

Pubmed is back! Obviously the DNS server issue has been resolved.

Nevertheless, I've created an account on Europe PMC and implemented my saved searches... better safe than sorry. Scientists are so dependent on NIH services, we should start to support more decentralized/federated infrastructure!

🧪

PubMed down .... Here's my post on what to do

Freaking out that PubMed is down. An incomprehensible act of vandalism. Trying not to join the dots to RFK's bizarre ideas.

It seems PubMed has disappeared. This is the digital equivalent of burning a library.

Two Weeks In The Death Spiral Of Cardiff University - Reading Time: 4 minutesBy an anonymous academic. Cover image by Adwitiya Pal On the second day of classes…

Two Weeks In The Death Spiral Of Cardiff University, by an anonymous academic.
This is truly shocking read.
The Cardiff executive board should hang their heads in shame.
voice.cymru/two-weeks-in...

Looking for an NIHR pre-doctoral fellowship opportunity in trials methodology? Feel free to contact me!

Even works in Claude.ai

The first consultation for the Evidence Synthesis Infrastructure Collaborative is now live. If you're interested in the future shape of global evidence synthesis do take a look and have your say!

evidencesynthesis.atlassian.net/wiki/spaces/...

@cvonbastian.bsky.social - I am aiming for BattleZone next.

Create an Asteroids-style game where I control a spaceship that moves with the arrow keys and fires lasers with the space bar. The game should start with multiple asteroids that move randomly across the screen. When a laser hits an asteroid, the asteroid should explode into two smaller asteroids if it's large enough, or disappear if it's small. If asteroids collide with each other, they should also shatter into smaller pieces. Additionally, if an asteroid collides with the spaceship, the game should end. When all asteroids are cleared, a new wave should appear, increasing the number of asteroids by one in each new wave. Also, add a score counter that increases when I destroy asteroids. Provide the full game as an HTML file with a link so that I can download and run in my browser, with all JavaScript included in the HTML file itself.

Challenged myself to write a single prompt for gpt4o that makes a game - came up with this after lots of iteration - now got a challenge going at work - who can make the best game from a single prompt (prompt in the alt text) #gamefromaprompt

I'm so sorry to hear this.

I was removed from my position as a Biomedical Librarian at the NIH Library by a 10 pm Saturday night email. I had worked there for 5.5 months.

Still figuring out what to say and how to say it, but feel free to get in touch.

Also would appreciate any leads on jobs or side gigs.

Love you all.

Elden Ring or Tenure-Track Professor? 1. You’ve invested an enormous amount of time and energy, and you still have no idea what is happening. 2. Everybody you meet speaks in half riddle...

"You want to stop, but the amount of time and energy you’ve put into this makes it hard to quit."

We need to listen to what Trump and Vance are telling us. This may have been the biggest moment for European security since 1991.

Have written about JD Vance’s Munich speech. In what may prove to be the biggest week for European security since 1991, it’s time Europe listened to what Trump and his Vice President are telling us. open.substack.com/pub/goodalla...