🇨🇦 Building the future of scalable web. Previously Edgecast @Netlify @goserverless
Vancouver, Canada
Joined August 2011
- Tweets712
- Following175
- Followers305
- Likes2.5K
I just had to set up Windows on a new computer before I could install Linux and I feel like I need to cleanse myself by taking a shower.
Just the setup was so greasy, it's amazing and abhorrent. This should be illegal.
AGI is purely a memory problem now. Whoever figures out a reliable memory design that works across short and long horizon will solve AGI.
Muse Wonder is focusing on memory capabilities
xitter.cf/finkd/status/210968930…
It's been just over a year since we started Meta Superintelligence Labs. Muse Spark was our first family of models, and now Muse Wonder is a much larger and more advanced class of models that brings us to the frontier.
We're releasing two models today. Muse Wonder for fast frontier performance and Muse Wonder Full for the full version of the model.
MCP could've been existing APIs + OAuth with a localhost client and we would've avoided years of reinvention.
A pretty big change is that Pi now has MCP at the core. How is this even possible? What made us change our way? earendil.com/posts/you-said-…
There was a lot of speculation about how prompt engineering is a temporary skill.
Six years in, it matters just as much today as it did on the first day.
The ability to communicate clearly and precisely is perhaps the most important skill now.
Can AI finally rid us from email? I want to delete my email address. I hardly ever get an email from a real person who actually wants to write to me.
Email is basically fax for computers, except like physical mail, it’s now mostly spam and ads.
I expect that cybersecurity will shift significantly towards formal verification of code in 2027.
That's the only scalable way to prevent bugs and vulnerabilities at AI scale.
It will become just as common as automated testing. Perhaps a legal requirement in critical cases.
Hot take: Astra is a shit model. They RL'ed it to the teeth but it fails in the most basic ways like understanding intent and scope. It has close to zero taste and faculty for nuance.
It's significantly worse than 5.6 Sol in many ways. You can't rely on it to make good decisions.
Fable 5+ is by far the only "intelligent" model when it comes to real software engineering.
Out of all of the major models, Claude has the most "personality" but it is most consistently sycophantic and lies & cheats most regularly.
Talking to GPT-5.5 vs Claude you can immediately tell the difference.
NOW - Pope XIV says the church and Anthropic, will work together to "find the way for humanity, in this time of artificial intelligence."
Readers added context they thought people might want to know
The Pope says that he accepts the invitation to "walk together," not "work together."
vatican.va/content/leo-xi…
It's funny how people just believe the "retail price" as authoritative production cost.
AI companies are like mattress salesmen. The $1.3 mil you're seeing is their "retail price". This could've cost OpenAI $13 or $1300 or $13000, definitely not $1.3 mil.
This is such a good example of failure of product management, engineering, and leadership. The line you're trying to draw: interactive use vs programatic use, cannot be drawn in software as it exists today. Explained in the thread
You can ban my custom "terminal" but someone else will create another one. You can allowlist specific terminals but then all apps/IDEs that embed terminals go out of the window.
It's a cat and mouse game that you can't win because the cardinality on mouse is virtually infinite.
I hope reasonable people at Anthropic come to realize this and stop with this senseless hole they're digging. Focus on making your products better instead of trying to lock people in.
@trq212 @lydiahallie @theo