Eben Upton

Unsure where to go for #CVPR2024 Thursday afternoon poster session? Try number 73, "Bézier Everywhere All at Once: Learning Drivable Lanes as Bézier Graphs" And to explore all the full text papers from CVPR in one place, with a huge range of annotation and exploration possibilities, check out http://app.drisk.ai

10

The case for GB Cloud The AI revolution will need massive computing capacity to enable it. In this paper Demos ‘ Elizabeth Seger with UKDayOne calls for the next government to invest in sovereign compute - GB Cloud - to develop cost effect and resilient compute. GB Cloud should be a public interest company offering discounted access to the NHS, public services and researchers and start-ups and could have parliamentary oversight and public consultation to ensure it is a trusted resource for the nation. The next will have to pick some targets to stimulate growth. This is a good bet. https://lnkd.in/eg28dipv

5

"AI is only one of many emerging technologies - from genome editing and 3D printing to a globally networked 'Internet of Things' - shaping a future unparalleled in human history in its promise *and* its peril. Are we up to the challenge this future presents? If not, how can we get there?"   Shannon Vallor asked this question in 2016 in the introduction to her timely book, Technology and the Virtues: A Philosophical Guide to a Future Worth Wanting. Eight years forward, and here we are, ChatGPT and all.   Vallor's contention is that, to meet the challenge, "we need to cultivate in ourselves, collectively, a special kind of moral character, one that expresses what I will call *the technomoral virtues.*" This puts her squarely in that ancient Greek tradition, starting with Socrates and full blown in Aristotle, that sees ethics as less about particular choices and more about the good life, "the kind of life worth choosing from among all the different ways we might live." This is the virtue tradition, and it's one that, while certainly undergoing a robust recovery in recent decades, is not the dominant ethical tradition in business or government. (That would be utilitarianism, more on that in future posts.)   While technology has shaped ethical thinking from the outset, we're facing something new, according to Vallor. So our ethics need an update, as it were. We now live in an ethical world where "our aggregated moral choices in technological contexts routinely impact the well-being of people on the other side of the planet, a staggering number of other species, and whole generations not yet born."   She's right about this. Just think, here on #EarthDay, about the staggering loss of biodiversity we've seen since 2016. Just think about the planetary thresholds we've transgressed. Just think about the #climateanxiety felt by billions from all corners of the world.   Aristotle could not have imagined such a reality. But maybe he, and Confucius, and the Buddha, and others who have thought about the good life can help us reimagine ethics for a new era. And maybe we can find new capacities, new excellences, to help us thrive in a technomoral world.   This is the challenge that Vallor has set for herself and for us. Let's see what kind of people we may yet become.   #ethics #technology #AI #virtue

2

With the UK general election coming soon, it's quite startling how embedded age has become as a key factor in voting intentions (albeit one that plays out within the context of the other variations within each constituency). Source of image (based on recent polling) and further analysis: https://lnkd.in/efhczst7 This has been the case for some time now: see e.g. this study of the 2019 general election - https://lnkd.in/e3VWaTFm The UK's a bit different in this regard from some other countries in which gender plays a bigger role - see this interesting analysis by John Burn-Murdoch in Jan 2024 - https://lnkd.in/e2ysFqrt That being so, it's interesting how the two major new policies suggested by the main contenders in the last day are age-related: Labour 'may' enfranchise 16-17 year olds and the Conservatives will reintroduce conscription fot 18 year olds. One important factor in UK elections has been that older people are much more likely to vote than younger. Recent legislation has also introduced requirements which have been criticised in some quarters as likely to suppress younger voters. The Electoral Commission's research indicates that in 2022 only 60% of 18-19 year olds and 74% of 25-34 year olds were even registered to vote, as opposed to 96% of people over 65 - https://lnkd.in/eFaGpfqN Footnote for anyone not familiar with the minor UK parties - -- Lib Dems are centrists. They're the rump of what used to be one of the two dominant parties until about a century ago, merged in the late twentieth century with some centrist splittists from the Labour party. Seemingly in a permanent state of too strong to die, too weak to win. -- SNP are the main Scottish independence party. All-conquering for over a decade north of the border but more threatened by Labour now. -- 'Greens' covers - I assume - the separate Green parties of England & Wales / Scotland. -- 'Others' presumably mainly lumps together the Welsh independence party (Plaid Cymru), some SNP splittists (Alba) and various Northern Ireland parties. -- Reform UK Party Limited is a right wing nationalist company (see https://lnkd.in/eVsbsEts) controlled by Nigel Farage. It's the latest incarnation of what has previously been known as UKIP (UK Independence Party) and the Brexit Party. Similar vibe to the VVD in NL, AfD in Germany, RN in France, FdI in Italy, Fidesz in Hungary, the modern GOP in the US and such like. -- Last time we had a general election in 2019, Farage's previous vehicle was similarly threatening to take votes off the Conservatives, but he stood down his people at the last minute and the Conservatives sailed home. He's said he won't be doing that again. We'll see.

25

28 Comments

Hello Marc Allera… The way/attitude of John in your Executive Team at EE twice now has been absolutely appalling. It is a real shame that #EE are treating Customers of nearly 30 years like this. I wonder how many other customers have been treated this way and have left EE? How come the quality of #CustomerService has fallen so much since combining with BT Group? Where has the great #CustServ that EE had gone? Out of interest, how many of my connections, followers etc had bad service too? Who can provide a good service today? Vodafone? O2 (Telefónica UK) Tesco Mobile Three UK giffgaff Neil Gregory, Julie Wilkin, Russell Johnstone, Sally Fuller, Christian Barrou Thrane, Chris Pile, Amit Sandal, Chris Bray, Sharan Gandhi, Mat Sears, Niamh O'Connor, Lee Frankham, Sean Cundy, Stephen Glasson #MarcAllera #EE #Vodafone #o2 #TescoMobile #Three #GiffGaff #CustomerEngagement #CustomerExperience

1

1 Comment

Here is what I did in my spare time this week. Subject is expert panels, mine and Sam Altman's favorite niche in GPT4, which, as Sam likes to say these days, "kinda sucks." Even so, the days of human expert panels are numbered. All that and more, including experts splaining hallucinations. My theme art style for the graphics this week is Art Deco. https://lnkd.in/eif68y7Q

1

2 Comments

As the Chief Technology Officer at a Health startup using AI, I see the introduction of the "Artificial Intelligence (Regulation) Bill" as a critical juncture for the technology sector. This proposed legislation offers a chance to establish a regulatory framework that not only protects the public and maintains ethical standards but also encourages innovation. The creation of an AI Authority could serve as a centralised body to ensure consistency and coherence across various sectors, effectively addressing the current regulatory fragmentation. Such regulation is crucial to balance rapid technological advancements with the need for transparency and accountability, ultimately fostering a more trustworthy AI ecosystem. This bill could set a precedent for global AI governance, positioning the UK as a leader in this vital area. Lord Holmes #aibill #ArtificialIntelligence #startups #ai #ukai #aijobs

12

As we increase our digital surface area (more services / smart devices / IoT), do we lose sovereignty over personal data? Here are some reflections from engaging with the fantastic panel last night... ❌ What it means to be forgotten is changing, as traces of our data remain embedded within LLMs, algorithms and AI models. Do we simply have to make peace with this? 🕵️‍♂️ Deepfakes are coming to take your lunch. Spoofing and phishing is getting smarter, preying on our nature. Some form of multi-factor authentication in the real world is soon going to be a reality. 🎛️ It's too simplistic to say consumers need to take greater responsibility for their data. (Even as we operate under the assumption that platforms remain accountable when we ask for certain data not to be collected.) 🧐 There is a growing education gap, consumer fatigue and (dare I say) apathy. Who hasn't been guilty of mindlessly accepting cookies? 🏛 I'm not sure governments or regulators are necessarily the answer. But then who? As always, innovation inspires more questions, and change brings us greater problems. Viewed through a philosophical lens, is having ownership and control over our privacy part of what makes us human? Thank you Picasso for holding an inspiring event at LSEG (London Stock Exchange Group).

12

3 Comments

One of the myths I hear about Digital Forensics Triage time and time again is that it creates "a risk we'll miss something". Used properly, it doesn't. In fact it achieves the opposite - it reduces the risk that you miss something until it's too late. We've seen lives saved and victims safeguarded as a result of sound triage, including triage using Cyacomb technology. Just in the last few days we heard of a case where our fast on-scene triage at a family home helped officers realise that someone other than the expected person was definitely offending enabling prompt safeguarding action. That was transformative to the investigation - which would otherwise have relied on reports from a lab weeks later. We have customers using Cyacomb Examiner in different ways depending on their circumstances and needs. Some of them are trading off different types of risk in quite sophisticated ways. None are accepting significant new risks and nor should they - when it comes to protecting children and catching those who seek to harm them we can't afford for things to fall through the cracks. Find out more here:

6

1 Comment

🤖 👀 not sure if this is publicity, fonzi-esque shark jumping, or the future... (I did check, it's not an April Fools) https://lnkd.in/eDQqNMz5 #CsuiteAI #artificialintelligence #AI

5

1 Comment

If we listen to the spin machines in Big Tech, LLMs are advancing rapidly towards AGI, perhaps with a year or so of slippage being mentioned occasionally. However, as this article points out, this may well not be true. "My brother and business partner, Addison Best, and I have been noticing the capabilities of GPT-4o declining. Yes, this is anecdotal, but the decline is pretty obvious." "For example, if I ask GPT-4o to put my affiliate links in certain spots of an article, it might mess this kind of simple task up. (It was working better before, but somehow the model’s abilities seem to fluctuate over time)" This is interesting. Is 4omnia (multimodal) meant to be better? Even OpenAI's graphs show that capability of 4o is only marginally better than 4T and in some cases worse. Much of the available data shows that LLM capabilities are plateauing very fast. "It seems that while GPT-4o is indeed super fast (it really is), its accuracy and understanding of what I’m asking it to do seems worse than even GPT-4 was." Possible reason suggested are "Companies have a huge financial incentive to lie and exaggerate about AI to get more eyeballs and funding." "Companies have been actually caught exaggerating claims during their AI demos (Google, OpenAI, Amazon, etc)" "The models seem to be getting slower (personal experience)" "While the number of model parameters goes up exponentially, performance doesn’t. (You can’t squeeze unlimited juice from a lemon just by squeezing harder)" And the final killer reason for decreasing LLM capability is "The problem is that the best models aren’t nearly energy efficient enough to give out for 20 bucks a month." What is your experience, is 4o better or worse than 4T, faster or slower? What is the empirical, experimental evidence? https://lnkd.in/ei26kKw4

45

31 Comments

I've been reading Prof. Bernard Porter's 2007 second edition of his mid-1960s PhD-thesis-turned-book 'Critics of Empire.' It covers the period from the mid 1890s to 1914 and focuses on British critics of British colonial / imperial activity in Africa. His focus is not on the pro-imperialists, the anti-imperialists or those apathetic about the whole thing, but rather on the 'critics' - by which he means people like John Hobson (1858 - 1940) and Ramsay MacDonald (1866 - 1937). To quote Porter: 'In rejecting Manchesterism and the doctrine of non-intervention, they found themselves impelled to admit the possible theoretical justification of imperialism; Hobson on grounds of international social efficiency, MacDonald on grounds of international moral responsibility. But both of them were quick to point out the ethical problems which a practical imperial policy involved; the danger of abuse, the absence of any means of practical arbitration between nations; and, in particular, the complex difficulties produced by the 'multiform' character of civilisation. Hobson appeared to be overwhelmed by these problems... MacDonald, however, was willing to compromise..." The young Porter disarmingly states in his original 1960s preface that his aim has been to unearth these controversies not to judge them, as that would require "a wide and detailed knowledge of the facts" which he does not possess. That humility makes it a fascinating book. There is also a lengthy and interesting introduction to the 2nd edition, written in 2007. He observes that the book first appeared at the height of the Vietnam War, "widely criticised at the time as an imperialist venture" though he notes that, when he was researching it, in the mid 1960s, it wasn't as big an issue in America and Britain as it became a few years later. The rest of the introduction addresses some developments and reflections since the 1960s, and notes that "even with a proper historian's caution against exaggerating superficial parallels" he is struck by the echoes of the debates in the 2000s about the US-led invasions and occupations of Iraq and Afghanistan. He expresses the hope that "new readers today will also be able to view it in this treble way: as a source of knowledge and understanding of the early 1900s (of course), of the 1960s, and for our present times." It certainly has some limitations, but it is a very interesting record of some of the mood and arguments of the times. And as the 2000s now fade into history themselves, the 'treble way' is now quadruple (think Brexit, Trump, Biden etc). The picture on the cover is a modern variant of an 1899 Punch cartoon featuring the imperial general Kitchener and (caricatured as a 17th century Puritan) CP Scott of Manchester Guardian fame.

11

4 Comments

Multi-species SNP chips are a game-changer, but there's a catch. These chips might miss important genetic variations, especially in exotic populations. When genotyping related organisms, they focus on diverged loci, ignoring common variation. Imagine studying a language only for new slang, missing the core vocabulary that connects everything. I hope that my recent paper could guide researchers to better design new chips to avoid increasing the ascertainment bias https://lnkd.in/g-5xTa66

26

8 Comments

What’s it like as a woman working in tech? There have been plenty of statistics and reports in recent years, but they don’t really tell the full story. Enter Kristi Coulter’s book “Exit Interview: The Life and Death of My Ambitious Career”. She talks about the highs and the lows of navigating a career at Amazon, and the way you can’t outwork being a woman in a male-led space. I’m sure plenty of women who’ve worked in the big tech world will find something in this memoir that resonates. #womenintech

34

2 Comments

For the past few weeks, I've become more interested in Quantum Theory and how scientists are trying to unify the theory of relativity with quantum theory. But not only that, I’m also intrigued by how quantum theory could be key for AI. What's great about all of this is how the current state of AI can help non-physicists like me understand these concepts. I wanted to share a conversation I had today with ChatGPT on this matter. https://lnkd.in/ghTpEjQG

3

1 Comment

Road signs are user interfaces and computer systems should accommodate them appropriately. Systems should be designed to accommodate the real world, instead of changing the real world to accommodate deficiencies in the systems. This should be true for place names like St Mary’s Walk or Martha’s Vineyard or people’s names like O’Toole or D’Sousa. https://lnkd.in/eZ7s-xNp Let’s not rename Westward Ho! or Ashby-de-la-Zouch, either. If systems can sensibly sort McDonald and MacDonald together, they can handle St Mary’s, St. Mary’s, St Marys and ST MARYS as variants. The US has a similar daft “discouragement” of apostrophes, but cannot explain why: https://lnkd.in/ekjG5N2r #uxfail #namingconvention #ucd

2

2 Comments