Joined July 2016
218 Photos and videos
Pinned Tweet
8 Feb 2025
I've been watching this game long enough. Time to play 👾
1
16
3,104
prediction: open AI will release a new model in less than 2 weeks source: gpt-5.5 is getting dumber by the minute
34
Jun 13
A new instance of "hey, we just found out you owe us €100 from 2013" from the Spanish authorities. Timeline: - November '23, I get sick. GP gives me the wrong medicine, and 7 days of sick leave, which is the default - I get worse, so I need to visit urgent care. Doctor there gives me the right medicine - I go back to GP, they extend the sick leave for another 7 days, but they forget to communicate that with the company that pays when you're sick (more on this in a bit) - May '26. Social Security sends a letter telling me I owe them €115. I spend a few hours trying to find out why, and then I discover it's because the GP forgot to communicate the extension. - Now I need to spend more time trying to get proof of the extension, which only my GP can provide for privacy reasons. The fact that I no longer live in Spain obviously make this more complex I was freelancing at the time, and many people will be surprised about a freelancer having sick leave and getting paid while being on sick leave. Some things to keep in mind to understand this better: - it's NOT free, and you have to pay 30.5 of your revenue before taxes for it, capped at €1500 per month. - there's a minimum on the amount you have to pay, so even if you don't make money you have to pay €200 (it was 300 in 2023) - the first 3 days of leave you don't get paid, then you get €17 per day. Note you still have to pay for at least the €200 minimum. So you pay a lot of money because the state "have you covered when you need it". Then you do everything right, and 3 years after you leave Spain you are told you owe €100, incorrectly, and you have to spend multiple hours worth way more than what you "owe", to explain they actually made a mistake and you don't owe anything. And you can definitely pay the "don't bother me" fee and forget about it, but in my experience that brings more and more requests like this over time.
1
2
80
Jun 13
This is an awful precedent, where you can use models based on your nationality. I hope this is just part of the power games Anthropic and the US government are playing, and will be part of the past soon, but in any case this should make EU leaders think about what we're doing with AI.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…
44
Jun 11
we should rethink PR reviews. It's no longer an async process, and it's not even "PR" anymore. It should be more like a dialog over a diff changing in real time. I'm not going to look at the whole set of changes and leave a bunch of comments. I want to ask as I'm reading the changes, and then the changes are updated in real time. Is this already part of an existing product? Any extensions (zed or vscode) I should try?
31
I did this workout again today and finished 45 seconds faster! It's a big improvement considering I couldn't beat my best time for the last 6 months. You can see the last two red dots dots clearly breaking the trend. Just having a clear target can make a huge difference.
I managed to cut 33 seconds on a workout that I couldn't improve for the last 6 months The key: go SLOWER My best time had an average of 5:50 per round, so I set out to do 5:30 per round. That meant 30" slower for the first round, but then I could keep the pace for the rest!
45
Every single time. If i tell codex/opus what to do and iterate on the change without looking at the code, I end up with a big mess. It's the reverse raptor effect.
22
Imperial units are just crazy and make no sense. But we all have to be grateful that we at least agreed on time units.
1
51
Daniel retweeted
i still think about this specialization is for insects
127
1,174
9,223
184,868
i posted this meme to reddit, just to see if there's any difference with X. I kid you not, it got downvoted because "it looks like a repost". So you can't use a meme template there 😂
sometimes numbers are scarier than they should
52
sometimes numbers are scarier than they should
87
This should be very basic, but: - cheap = abundant - expensive = scarce Many politicians claim they'll "make X cheaper", then proceed to implement policies that make X even more scarce. Share this with your politicians. Maybe they didn't know about abundant and scarce goods.
29
Daniel retweeted
Add Flint to your lock screen. Create voice note without waiting.
1
1
120
If you haven't used Hermes yet, you have no excuse anymore.
We are excited to join Nvidia's Nemotron Coalition of leading AI labs working together to advance open frontier foundation models. To celebrate we have partnered with @nvidia and @nebiustf to provide 2 free weeks of the new Nemotron 3 Ultra model on the Nous Portal!
1
152
confirmed, no clue what I'm doing. random post: 24k views, 150 likes, 20 comments thoughtful post: 72 views, 1 like, 1 comment
2
77
Hot take: B2C Intelligent apps won’t exist in a few years. We’ll see them the same way we see a PDA. It was a key step to reach cellphones, but they didn’t last very long. They’ll be replaced by personal agents (i.e. Hermes, OpenClaw, etc). Those agents will still interact with 3rd party services, but those services won’t provide the intelligence layer. The two main drivers of this replacement: - An intelligent app can’t compete with the level of personalization of an agent - With your agent you can decide how much intelligence to spend on each task What will 3rd party apps do then? They’ll manage the data and social layers. The data layer is both collecting data to feed personal agents, and presenting agent’s output inside a domain specific context. Data collection is both sensor- and human-driven. Effortless is the key here. Examples are those apps that estimate calories from a picture, voice notes with super powers, or workout trackers that count reps based on video and smartwatch sensor data. Presenting the output in a context means showing the output where it is actionable. For example, your agent proposes a workout based on state of the art scientific research, your fitness goals, previous workouts, but also stuff like sleep score the previous night, or even the weather. The workout is exactly what you need. But it’s almost useless if you get it in an email. You might be thinking that your agent will build the UI to present its output in that way. While this is true in many cases, there are two important aspects to take into account: - domain knowledge > personalization: while your agent can tailor things to you, some times we need to tailor things to the domain. In some cases, knowing a lot about the domain might be more relevant than knowing a lot about you. - the social layer: sharing content with others from a common app has significantly less friction than having people install apps custom built for someone else. Note that the “domain knowledge > personalization” is the reason this only applies to B2C. For B2B, having an intelligent app that knows more about “businesses like yours” might still be more relevant than knowing specifically about “your business”.
2
1
122

Jun 5
Software platforms are going to be rebuilt for agent-first.
41
My usage limits just got reset overnight 😍 Let's see how they last this week
after a week form switching to Codex from Claude, my take: ✅ Codex with 5.5 feels a bit faster than Opus 4.7 (haven't tried 4.8) ✅ Codex feels more intelligent than Opus when building things, but is not as good at understanding ❌ Weekly limits are significantly lower
40
I definitely don't understand the algorithm 😂 At least it's refreshing to have more than 10 people read something I posted and actually interact with them. Tomorrow I'll post a hot take much more interesting than Codex vs Claude. We'll see how that goes.
1
2
204
after a week form switching to Codex from Claude, my take: ✅ Codex with 5.5 feels a bit faster than Opus 4.7 (haven't tried 4.8) ✅ Codex feels more intelligent than Opus when building things, but is not as good at understanding ❌ Weekly limits are significantly lower
23
3
150
24,644
limits are a problem tbh I'm about to reach the limit, either today or tomorrow. That's 4 days and not very intense ones. I'd need a 30% increase in limits, but the next plan is a 5x increase in price. All in all, I think Claude is a better deal at the moment
4
1
5
2,149