Reforms of science funding in the UK aim to “do fewer things better”.
Clearer priorities could strengthen impact, but uncertain timelines and budget shifts may slow research momentum, especially for early-career scientists.
🔗 https://www.nature.com/articles/d41586-026-00447-6
#ScienceFunding #ResearchPolicy #UK #EarlyCareer #ResearchEcosystem
That Letter from UKRI
I only have time for a quick post today but I think it’s important to comment on the very feeble open letter circulated (yesterday) to “the research and innovation community” by the Chief Executioner Executive of UKRI. I think it’s feeble because it seems to have been intended to clarify what is going on, but does nothing of the sort. In fact, to me, it reads like it was written by someone who doesn’t know what he is doing.
The letter basically tells researchers working in areas outside the STFC remit (i.e. in anything except particle physics, astronomy and nuclear physics) not to worry because it’s only STFC that will suffer. This is the “explanation”:
In order to remain sustainable, STFC must make significant cumulative savings: a decrease of £162 million relative to our forecasts for their operational costs. The £162 million is the total net reduction in STFC’s annual costs that they must achieve by the end of the 2029 and 2030 financial year. It is not a £162 million saving in each year of the current SR period. Instead, STFC needs to reshape its cost base over the whole SR period so that their budget is balanced by 2029 and 2030 and key facilities are funded properly and sustainably.
That is not the situation at other councils and we do not anticipate equivalent measures will be necessary outside of STFC.
One of the problems with this logic is that a huge slice of STFC’s budget is spent on facilities that support science outside STFC’s scientific remit. The Diamond Light Source, for example, which has annual running costs of almost £70 million caters largely to the EPSRC and BBSRC communities. It makes no sense to me to require particle physics, astronomy and nuclear physics reseachers to bear the entire consequences of cost overruns at this facility when other communities benefit from it.
I’m sure the UKRI Chief Executive knows this, so it must have been a deliberate decision to wield the axe in this way. In other words it’s a conscious downgrade of particle physics, astronomy and nuclear physics. In the new regime, these are less important than any other branch of scientific research.
I’m out of it now, but I always felt that STFC should never have been set up as a research council. It should have been a service organisation, as its title – the Science and Technology Facilities Council – suggests. When STFC was created, back in 2007, funding for particle physics, astronomy and nuclear physics research as opposed to facilities should have been administered by EPSRC. Whether intentionally or not, the current arrangements make these areas of fundamental physics exceptionally vulnerable. We saw the consequences of that back in 2007/8 and it is happening again.
📢𝗣𝗜𝘀 & 𝗖𝗼𝗻𝘀𝗼𝗿𝘁𝗶𝗮 𝗳𝗼𝗿 𝗼𝗰𝗲𝗮𝗻 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵!
Apply to the #BelmontForum for funding to advance ocean biodiversity & ecosystems
📅05 Mar 2026
🏷️Know someone that should apply? Like or share
🔗 https://rxn.mbp-rnc.com/25adm?utm_source=Mastodon&custom_day=Thursday
#ResearchFunding
#FundingOpportunity
#ScienceFunding
#ResearchSupport
US congress reverses Trump’s budget cuts for science.
The U.S. Is Funding Fewer Grants in Every Area of Science and Medicine
https://www.nytimes.com/interactive/2025/12/02/upshot/trump-science-funding-cuts.html
Transformer components such as non-linear activations and normalization are inherently non-injective, suggesting that different inputs could map to the same output and prevent exact recovery of the input from a model's representations. In this paper, we challenge this view. First, we prove mathematically that transformer language models mapping discrete input sequences to their corresponding sequence of continuous representations are injective and therefore lossless, a property established at initialization and preserved during training. Second, we confirm this result empirically through billions of collision tests on six state-of-the-art language models, and observe no collisions. Third, we operationalize injectivity: we introduce SipIt, the first algorithm that provably and efficiently reconstructs the exact input text from hidden activations, establishing linear-time guarantees and demonstrating exact invertibility in practice. Overall, our work establishes injectivity as a fundamental and exploitable property of language models, with direct implications for transparency, interpretability, and safe deployment.
LeidenForce
In the Dark
Matthew B. Plutschack
WYM Mod
Georg Weissenbacher
Hacker News
Steven Saus
N-gated Hacker News
Ars Technica News