There is absolutely no other way to deal with this. AI assisted vulnerability discovery is, long term, a blessing. The bugs were always there, we are just finding them at a quicker rate. The next year or so is going to be painful but the end result is more resilient systems.
The only way out is through.
They're calling it the Bugpocalypse. LLMs are being used to find long-hidden cracks in the software infrastructure, and it's happening at a furious rate. 3% of all compromises in computing history happened in the last week, claims one source? I believe it.
There are people saying public access to the frontier models ought to be locked out, or heavily guardrailed, over this. The problem with that idea can be summed up in one sentence: when lockpicks are outlawed, only outlaws will have lockpicks.
The lockpicks will be Chinese open-weight models, if nothing else. There's no way to stop those from becoming ubiquitous.
There are two futures in front of us. One in which only the bad guys have lockpicks, and security holes go undiscovered until they're exploited. Another in which everybody has lockpicks and is aggressively red-teaming their own software so it gets fixed before the holes are exploited.
The corpus of human code is huge but finite. The number of defects in it has to be finite too. We can fix those defects now; we have a million friendly robots to do it with.
The only way out is through.
Rob retweeted
The only way out is through.
They're calling it the Bugpocalypse. LLMs are being used to find long-hidden cracks in the software infrastructure, and it's happening at a furious rate. 3% of all compromises in computing history happened in the last week, claims one source? I believe it.
There are people saying public access to the frontier models ought to be locked out, or heavily guardrailed, over this. The problem with that idea can be summed up in one sentence: when lockpicks are outlawed, only outlaws will have lockpicks.
The lockpicks will be Chinese open-weight models, if nothing else. There's no way to stop those from becoming ubiquitous.
There are two futures in front of us. One in which only the bad guys have lockpicks, and security holes go undiscovered until they're exploited. Another in which everybody has lockpicks and is aggressively red-teaming their own software so it gets fixed before the holes are exploited.
The corpus of human code is huge but finite. The number of defects in it has to be finite too. We can fix those defects now; we have a million friendly robots to do it with.
The only way out is through.
Rob retweeted
Need more GPUs literally today - anyone expecting their usage to go down over the weekend and happy for me to short term sublease?
I think @GothOnABrompton s X account has been hacked.
It is sending DMs with phishing links to compromise other people's social accounts.
If anyone knows her IRL, please let her know, and don't click on links in any DMs you get.
A domain name is for life, not just one election.
The dangers of letting them expire when old social posts still point at them.
Proud to launch A Radical Labour Vision for the 21st Century - my election manifesto - this morning. Read it here: andy4leader.co.uk/manifesto
Rob retweeted
new post: collecting favours
sunilpai.dev/posts/collectin…
a post on the idea of "reputation". tried writing it a few times as an abstract thing and I hated it, so instead I just wrote about my dad.
The predicate here is that local alternatives will always be worse.
Is this really true?
If it is, how do we fix that?
The Palantir question is the most interesting issue in UK tech today.
To what extent are we willing to trade quality for sovereignty?
Any local alternative will, without doubt, be worse than Palantir.
This week we've seen the French kick off about having to switch from Teams to a local alternative (and how terrible it's been).
The difference for us is that switching to a worse alternative could mean fewer free hospital beds and fewer police officers on the street.
Is that a sacrifice we're willing to make?
Personally I'd love to see @louismosley leave Palantir and raise $500m to build a European alternative. This would becone one of the fastest-growing tech companies in Europe overnight.
Say what you like, getting @Grady_Booch to defend the term "artificial intelligence" is a measurable tech Overton Window movement.
It’s artificial intelligence, Sundar.
And in the history of computing, this puerile attempt to rename it simply because of the simpleminded whims of an incompetent and corrupt narcissist will be seen as a failure of leadership, an abandonment of intellectual integrity, and an embarrassment for any who join with him.
This on tweet is everything wrong with our institutions in the UK. It seems like @britishmuseum have failed to model visitor flow, oversold tickets and just hand people this letter denying entry after they have travelled and queued. It is so disresepectful and cowardly.
When this happens, everything possible should be done to accommodate people, extending hours etc.
If this is not possible then the Senior Management responsible for the fiasco should be out walking the lines, explaining their failure to people who have paid and made arrangements, and doing everything possible to make it right on the day.
How did we allow organisations to start treating us so badly?
Waited since I was 7 years old to see the Bayeux Tapestry. Got in an hour-long online queue to spend £50 on tickets. Got there ahead of time. Stood in a queue for 20 minutes. Then was told halfway through the queue that the gallery was closing.
The @britishmuseum stinks 😭
Not great that page existed, although we don't know the context, but it is very encouraging that it got fixed and we have a Minster who understands why it was off.
I've met a few of the @i_dot_ai folks and what they are doing will make public services better if they get traction.
Being environmentally conscious and keeping an eye on running costs is, of course, important - but so is encouraging AI adoption and scaling up its use across the public sector.
This advice has been updated and we are reviewing similar pages to reflect the fullness of our ambition for the UK 🇬🇧
ai.gov.uk/knowledge-hub/how-…
Cc @AlexBerish
Rob retweeted
Forest City is the only project that will make the actual housing cheap rather than piling MORE debt on to young adults. It can be done. We’ve done it before. We just need to break the developer model stranglehold.
Let's be honest.
A lot of young people have stopped believing they'll ever own their own home.
That's what happens when rents are high, house prices are high, and saving for a deposit feels impossible.
My Labour government is stepping in to help.
With Your First Home, first-time buyers will be able to buy a new-build home with a deposit of just 2.5%.
Because hard work and saving should count for something.
And because the dream of owning your own home shouldn't belong only to those with help from the bank of Mum and Dad.
Rob retweeted
Replying to @shivmalik
There is an understanding between the developers and government that everything possible must be done to keep prices high. New supply must only ever trickle. The market price must never return to affordable to normal people. This help to buy bullshit is just bread and circuses.
Rob retweeted
Why doesn’t Britain have as many trams as France and Germany?
Answer: Cost.
We pay roughly twice as much to build a kilometre of track. (And some projects cost a lot more!)
Why is it so expensive? One challenge is the cost of moving wires and pipes.
Our kit is older, badly mapped, and water/power companies get to pass the cost of utility diversions to the taxpayer when we try to build trams.
We also need to move more because we have deeper track-beds.
Today, I visited Coventry where they are trying a new approach known as ‘Universal Track’.
It only goes 30cm (I.e. a school ruler’s length) deep, unlike existing approaches that go 60-90cm deep.
That’s above most utilities in most place. It also uses less concrete.
They reckon it’s much cheaper as a result. Potentially half the cost of a normal track.
It was originally invented for their smaller ‘Very Light Rail’, but they’ve since learnt it’s strong enough to take full-size trams.
They hope to have it commercially available in two years.
So @typesafeai Jev isn't conversational, but if have all your transcripts, and defined goals for all of your agents, it becomes super cheap to go back and score every single interaction. Then target the best and worst outcomes for some detailed LLM review love. Watch this space.
Rob retweeted
We can do MUCH better than this.
There is a problem here. Engineers that always pushed the envelope, now feel they absolutely must maintain that differential.
Not quite as simple as FOMO, but is pretty close: "I (and other people) can do this at light speed, therefore, I must now do all things at light speed".
Don't get this attitude. Explicitly asking someone to hand write code from scratch is fine.
You don't, they ask if they can use common workflow, you say yes, but then terminate the interview for it.
Scoring *how* they prompt Claude and drive it is a hugely valuable predictor.
I'm not a human factors engineer, but it is blindingly obvious to me that the square of carpet that forms the ramp needs to be a different colour to fix this.
We are very careful about where we put our feet.
The yellow border is literally distracting users *away* from the man-trap in their path.

























