Ali (@its_AliAli) header
Ali (@its_AliAli) profile picture

technology enthusiast | co-founder @ RapportCue. sharing what impresses me in tech and business.

United States
Joined January 2022
Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)Photo by Ali (@its_AliAli)
I think we should bring back the whiteboard. This is getting out of hand.
How easy is it to cheat on an AI interviewer, really? Our CEO - @rvivek thinks he’s got it figured out. Let's find out 🍿
2
83
Why does “I found this using Claude” sound so much more intellectual and sophisticated than “I found this on ChatGPT or Gemini”? Is this the rise of bougie AI?
1
2
45
Now that we can build almost anything, what to build oh what to build? For it to be worth it, it has to make you go “wow”
1
56
You can technically search LinkedIn by ethnicity. This is a fascinating byproduct of a search algorithm that clusters people by graph networks and performs semantic mapping between intent-based search terms and user profiles.
2
33
From October, a new wave of cybersecurity influencers incoming …
1
38
It is not adding up… I keep thinking about how smart these models supposedly are and how the frontier labs actually operate. If you literally control all the compute and super intelligence you shouldnt be playing normal business games. You wouldnt just be selling subscriptions and api tokens. You would use the model to find an unlimited money glitch in the economy. Literally reward hacking your way into billions, even trillions.
1
59
Scary how relatable this is
peak generational evolution 💀
1
28
Why can’t I get this when I ask chatgpt to write something.
1
36
Something like that …
This Post is from a suspended account. Learn more
2
50
Billionaire is the new Millionaire. Lockin fellas…
1
39
Just some late night thoughts. Tokens are the work currency of LLMs. What’s the human equivalent? When someone writes an essay, every sentence is distilled from years of experience, memory, intuition, and thought. A life compressed into words on a page. The output is slower yes, but the underlying computation is far richer and the network it comes from is denser. Human tokens are expensive, but they’re still the highest-quality tokens of work we know how to produce. Even in today’s AI crazed world. This raises a lot of questions for me, probably you if you are still reading. Should we start valuing work in terms of tokens? Not quantity, but quality. Is the answer to AGI an even more general-purpose model that can do everything? A model with “life experience”. Instead of expert or specialist models working together. Are we heading toward a future that yearns for human work? A future where entirely human-coded platforms exist, and their biggest selling point is simply that they were built by humans. Will we see “100% human” ads the way food advertises “organic”? You can already see this in creative spaces. Music. Writing. Art. Films. Hell, even on YouTube videos. Do we end up with two separate markets? One for machine work and one for human work? And if we do, what becomes more valuable: the output itself, or knowing a human was behind it?
2
53
Textbooks … more day clay tablets
1
27
DRINK FROM THIS: Friday, April 10. A few stories from this week. 181. That's how many working Firefox exploits Claude Mythos produced in testing. Opus 4.6 managed 2. Anthropic published the evals this week. Anthropic published the evaluation results this week. SWE-bench Pro: 53.4% to 77.8%. Terminal-Bench 2.0: 65.4% to 82%. Big jumps. But the exploit numbers are what shut down the release.
1
2
54
I can't stop thinking about: We're running out of benchmarks. METR's Time Horizon suite, one of the more rigorous attempts to measure autonomous AI capabilities, is being saturated. Frontier models can complete nearly every task in it.
1
1
59
The estimate: by mid-2027, no benchmark from 2026 or earlier will be able to rule out dangerous capabilities from frontier AI systems. We're moving faster than our ability to measure how fast we're moving. Also. Meta used 60 trillion tokens internally last month. All the books ever published total about 20 trillion tokens. Three times that. In thirty days. Token spend is easy to game and more tokens doesn't mean more value. But that number keeps landing on me.
17
I have a graveyard of Twitter bookmarks. Things saved to "read properly later." Later never comes. Forty more on top. Like a fire hose you can never shut off.
1
1
32
Wildcards at the end. Things I can't stop thinking about. No criteria. Just whatever won't leave me alone.
1
26
I'm calling it Drink From This. Two or three stories a week, ranked by actual impact. Ends with whatever I can't stop thinking about. The hose you can actually drink from. Wednesdays and Fridays. First one this Friday.
1
17