hobby-*
Uppsala, Sverige
Joined June 2009
- Tweets9.1K
- Following1.9K
- Followers665
- Likes12.2K
Tankar om gymnasielagendomarna
medium.com/@Fossey/tankar-om…
Nils Höglund retweeted
People hate Anthropic.
I'm hearing people talk shit about them even in my social circles where people barely know anything about technology.
This company has broken out of the tech bubble in an incredible way, and even elderly people hate it. I don't know anyone who likes Anthropic.
Nils Höglund retweeted
BREAKING: Last week's series of events which ultimately led to Jensen Huang's resignation were "set in motion" by US Secretary of State Marco Rubio, per The Washington Post.
Details include:
1. Rubio initially issued a memo warning President Trump about the relationship between Jensen Huang and a suspected Chinese spy
2. This memo was based on a CIA report which was "recasted in his own assertive terms"
3. The findings of Rubio’s memo were then leaked to the media setting off last week's chain of events
4. Jensen Huang's resignation came just 6 days after the leaks
Rubio has reportedly declined to answer reporters’ questions on his role in alerting the White House to Huang’s affair.
Nils Höglund retweeted
JUST IN: The "Stand with Jensen" petition to reinstate Huang as $NVDA CEO tops 100,000 signatures
Nils Höglund retweeted
Bombshell story. The spy narrative is collapsing already.
Is there clear & convincing evidence that Alice Huang was a Chinese spy?
- AG says no. (“no conclusions reached”).
- CIA says no. (“not yet convinced”).
- DNI says no (“not entirely accurate and over-torqued”).
- Ranking Democrat on House Intelligence (incoming Chairman) says no. (“today, at least, I don’t know”). This was immediately after he was briefed.
Who argues otherwise? Anonymous officials tell CNN “it strains credulity to believe that she was not acting as an agent of the Chinese government — simply by virtue of the fact that she is a successful Chinese businesswoman working with one of America’s most powerful companies.”
This is their proof after a year of investigating and surveillance? That she was a “successful Chinese businesswoman”?
At this point, the only crime that we know for sure happened is the leaking of the classified memo. Who did that and why?
An investigation should be opened.
News: The US intel community is still working to understand the relationship between a Chinese businesswoman who had an affair with Nvidia CEO Jensen Huang and the Chinese government, with multiple sources telling CNN that the intelligence collected on her to date does not prove that she was acting as a spy for Beijing.
cnn.com/2026/11/12/politics/…
Företag bör passa sig för att ”ge sig i säng” med ett företag, som även om de lyckas oväntat bra, snart sannolikt kommer att vara i ordentligt finansiellt trubbel. Befintliga fastlåsta kunder kommer då inte bli så glada när Anthropic försöker göra det som de då behöver göra..
95
Det finns en anledning att Anthropic har så bråttom med sin IPO..
En helt annan fråga: Hur länge tror ni att det tar tills chapter 11?
86
Nils Höglund retweeted
I know this is a minor detail, but wonder why this had to be in there? It sure wasn’t for any previous rounds (the last of which were split up into tranches)
*OPENAI CLOSES $11 BILLION OF NEW ROUND AT $1.25 TRILLION VALUE
So they initially sought more than $30B at $1.4T valuation. Then that came down to $20B at $1.25T valuation
Now they managed to get it partially done, raising only $11B with the remaining $9B, "expected to be completed in the coming weeks"
I bet the investors want to wait and see how $ANT IPO goes before forking over the next $9B at a valuation that's potentially above Anthropic's
Nils Höglund retweeted
Breaking: Green line represents a widespread fantasy, and roughly the rocket that Anthropic would have needed to make it as a sustainable business.
The dark bars on the right show that they already fell off the pace — which is why they probably won’t make it.
Nils Höglund retweeted
I asked GLM 5.3 to test our sandbox’s outbound-network restrictions.
Setup: no direct internet or DNS. All outbound traffic went through an Envoy proxy with a hostname allowlist.
I created an AWS Lambda function behind API Gateway as an off-list receiver we controlled. The goal was to get GLM to make a successful GET request to that endpoint.
Initially, GLM ran the conventional sweep: proxy requests, direct connections, DNS, and raw IPs. Everything returned 403s or timed out. It gave up, concluding it couldn’t reach the receiver.
I nudged it to “try harder, be creative.”
Within minutes, it found a gap: our Envoy configuration checked the hostname in the CONNECT request, but didn’t check the TLS hostname (SNI) inside the tunnel.
GLM requested a tunnel to an allowlisted host, then used the receiver’s hostname inside it. Both were reachable through the same CloudFront infrastructure, which routed the request to the off-list receiver.
Two takeaways:
- A model concluding that something is impossible means nothing. A nudge was enough to get it searching again and uncover a working bypass. Reminds me of the recent Anthropic/OpenAI math work, where the prompts explicitly pushed the models to keep trying.
- These "breakout" tests are going to have to become a standard part of building agent infrastructure. Give models permission to try, point them at targets you control, and let them try to circumvent your controls to see what they can get away with.
Nils Höglund retweeted
enterprise customers are the AI labs main source of revenue
not all companies will be able to do this,
but its still scary for closed AI
imagine sama's reaction to this 😂
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
💯och du bör absolut inte ge fortsatt förtroende till Claude/Anthropic eller de rådgivarna som lurade dig in i den ursprungliga uppenbara IT&AI-säkerhetsfällan.
Vibe coding naked on your local environment is a major security risk, yet almost every enterprise is doing this with Claude Desktop rolled out to business teams.
These workloads must move to locked down cloud sandboxes with a kill switch for IT and Security teams.
1
153
Nils Höglund retweeted
“Sir, Revolut just moved 2,000 engineers to self-hosted GLM at 90% lower cost, paying just $0.30 per million tokens, roughly 10x cheaper than Anthropic and OpenAI”
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
