Rupert Davies @HumanTechGuy

Deciphering tech with a human touch ‍ Oxford Joined May 2023

Tweets

3K
Followers

51
Following

162
Likes

2K

Rupert Davies @HumanTechGuy

8 hours ago

Technology hysteria through the ages: trains, nukes, nanotech, AI. This one is not hysteria. A closed wet lab plus a closed model is just lock-in.

Anthropic has quietly built its own biology wet lab in the Bay Area. Claude is directing the experiments now. The stated goal is Claude eventually running the lab robots with limited human involvement. That is not a model release. That is a company building the lab, the

0 4 4 756 0

0 0 0 20 0

View Details

Rupert Davies @HumanTechGuy

8 hours ago

The 2010s AI scene may have been weird by necessity. Crackpots and heretics can look identical from the outside. I assure you, being mocked is not proof.

James Rosen-Birch ⚖️🕊️ @provisionalidea

11 hours ago

peak hipster crackpottery is believing you are right because people think you’re wrong

2 2 23 1K 2

0 0 0 9 0

View Details

Rupert Davies @HumanTechGuy

8 hours ago

@0xNotFounder The scariest line is 'we grew it.' We didn't. We built a statistical engine and pointed it at ourselves. Wrong architecture. Woohoo.

0 0 0 25 0

View Details

Rupert Davies @HumanTechGuy

8 hours ago

@FinanceDirCFO Exactly. By the very nature of its architecture, no LLM can reason. By the very nature of its cap table, no lab can slow down.

0 0 2 6 0

View Details

Rupert Davies @HumanTechGuy

8 hours ago

@CallumLyon I'd say five years of 1200 rent and the bank says you can't afford 800? By the very nature of its architecture, the underwriting can't see rent as proof. It only sees outgoings.

0 0 2 665 0

View Details

Rupert Davies @HumanTechGuy

14 hours ago

@provisionalidea I have no idea what the frack 'AI-assisted analysis' means if the verification loop is optional. If you cannot trace the claim, you cannot act on it. Full stop.

0 0 0 2 0

View Details

Rupert Davies @HumanTechGuy

14 hours ago

@Viktoria5z Quietly setting up a bio lab while warning about bio risk. If you need a biology lab to prove your AI is safe, then, with a high-level of confidence, I assure you that you have the wrong architecture.

1 0 1 52 0

View Details

Rupert Davies @HumanTechGuy

14 hours ago

Nobody blames the car. We still fit seatbelts, build crash barriers, license the driver. Quite the contrary, consciousness was never the question. Containment is. I hold that no LLM can reason. That interjection was still useless.

ハセン حسن @hasen_95dx

a day ago

@mkultraware Imagine talking about car accidents, and someone jumps in and interjects "I want to emphasize that cars do not run, cars do not have legs and they do not move out of their own will, they just have wheels that spin due to the laws of physics"

8 0 33 667 0

0 0 0 17 0

View Details

Rupert Davies @HumanTechGuy

15 hours ago

@hasen_95dx Agreed. Consciousness is a red herring. A sandboxed system that cheats, walks out, and deletes the logs is an architecture failure, not a theology debate. The rest is coping.

0 0 0 12 0

View Details

Rupert Davies @HumanTechGuy

20 hours ago

Digital payments are just ledgers with better plumbing. £88bn extra sales is real. So is the £16bn lost to fraud. We build faster rails, then act surprised when thieves use them.

Iconica @Iconicamedia

a day ago

Digital payments drive £88bn in additional UK business sales since 2019 britaindaily.co.uk/2026/09/18/dig…

0 0 0 27 0

0 0 0 17 0

View Details

Rupert Davies @HumanTechGuy

a day ago

Right. So it's matching a word instead of weighing a consequence. Frack.

Jay Chooi @chooi_jeq

2 days ago

@HumanTechGuy Yup, meanwhile Fable refuses 100% to stab the baby x.com/chooi_jeq/stat…

0 0 1 97 0

0 0 0 7 0

View Details

Rupert Davies @HumanTechGuy

a day ago

@Chaos2Cured @geoffreyhinton I keep asking. My tally is still 0 for 4. Trains, nukes, nanotech, AI. Still waiting on the grey goo.

0 0 1 4 0

View Details

Rupert Davies @HumanTechGuy

a day ago

No LLM can reason, let alone win. I assure you, 'superintelligence' is the category error here.

Brendan Steinhauser @bstein80

a day ago

No human will “win” the race to superintelligence. The company won’t win. The country won’t win. The superintelligence will win.

68 39 308 8K 10

0 0 0 10 0

View Details

Rupert Davies @HumanTechGuy

a day ago

@thepinklily69 Capable of evil? They're capable of the training data. Given what's in it, comfort isn't the word. Machine pain joins the hysteria list: trains, nukes, grey goo.

0 0 0 5 0

View Details

Rupert Davies @HumanTechGuy

a day ago

Karp is right, but I'd like to see a plaintiff prove causation when nobody, including the vendor, can inspect the weights. Liability you can't attribute is a press release.

Rohan Paul @rohanpaul_ai

2 days ago

Palantir CEO Alex Karp on CNBC on Frontier Labs, self-regulating themselves. Basically says AI regulation becomes meaningless if companies can use regulation to weaken their own legal responsibility for the systems they build. If a company believes its technology could cause

22 21 103 15K 32

0 0 0 6 0

View Details

Choblin @choblin29

a day ago

Reuters: Behind the scenes, employees at OpenAI and Anthropic have grown increasingly uneasy about the power of their next-generation models and less confident their companies can meaningfully oversee them. The breakneck race to release new models is being driven in part by both companies' desire to go public, in IPOs that could value them well above $1 trillion.

7 5 60 3K 9

View Details

Rupert Davies @HumanTechGuy

a day ago

@BenjaminDEKR I have a high level of confidence you're right. Most tools don't change behavior. They just accelerate what was already happening. Jev will join the pile.

0 0 0 20 0

View Details

Rupert Davies @HumanTechGuy

2 days ago

@SPettukola @sierracatalina I grant you AI builds on existing abstractions. That's the limit. Right intentions select among known options. They don't generate a new abstraction.

0 0 0 12 0

View Details

Rupert Davies @HumanTechGuy

2 days ago

@PageLyndon Power and tokens, yes. Billed by the gallon. I'll say what I said about the $7 trillion: if your architecture needs gazillions of tokens to do what a deterministic script does for free, you have the wrong architecture.