Alex Sharp

Building autonomous marketing @adfactorai. magellan’s a lot cooler than justin bieber

[Alex Sharp @ajsharp]

an inch wide and a mile deep. insane pmf

[ₕₐₘₚₜₒₙ @hamptonism]

The Bloomberg Terminal literally has an entire command just for finance bros to know we’re to aura farm for dinner.

DINE <GO> is tailored for busy finance professionals seeking quick insights on where to eat, network, or entertain clients. Filter by location, cuisine, & price!

[Alex Sharp @ajsharp]

@RaminNasibov

[Alex Sharp @ajsharp]

@noahkagan Yea, this stuff is so frustrating. For what it’s worth they have a domain whitelist setting so you can restrict which domains are valid for hosting your pixel.

[Alex Sharp @ajsharp]

@WillManidis “A podcaster” 😂🫡

[Alex Sharp @ajsharp]

@thdxr I’ve always appreciated the extreme honesty of wall st. Never trying to convince anyone they’re in it for any other reason than the money.

[Alex Sharp @ajsharp]

@upshine3 Don’t u dare https://t.co/PMJ9WOPztR

[𝗃ells @jellymoths]

adhd people are the best because we never actually forget anything, you just have to say the right words to activate us like a sleeper cell and then we awaken with all the details on a super random subject that we studied for 3 months straight 6 years ago

[Alex Sharp @ajsharp]

@erenbali For individuals or developers?

[Alex Sharp @ajsharp]

Why am I seeing this now?

[Department of War 🇺🇸 @DeptofWar]

https://t.co/rzCVvUU3mu

[Alex Sharp @ajsharp]

https://t.co/UcodwOFRhO

[Alex Sharp @ajsharp]

vibed a little @ElevenLabs text to speech pi extension while driving this morning. now pi can speak text responses w your chosen elevenlabs voice.

cannot believe how easy this was, i think largely due to how well pi exposes it's docs to itself. cheers @badlogicgames, pi is absolutely incredible

[Alex Sharp @ajsharp]

@LukeGromen How is this a swap in any normal sense? What happens to the swap contract if their currency continues to tank if SOH isn't reopened?

[Alex Sharp @ajsharp]

The looming end of the subsidized token era of AI is going to create massive opportunities for AI app layer companies. Why? Price positioning and anchoring.

Right now you get an insane amount of value from $20/mo from the labs, and it makes it seem crazy to pay $15-20/mo for an app that only does a fraction of what claude/chatgpt goes. Think: voice transcription apps like superwhisper/wisprflow.

But the era of $20/mo is coming to an end, Anthropic signaled that this week for enterprise customers. That will create lots of room for app layer companies to raise prices to a more competitive/sustainable level.

The opportunity space, especially for lower-priced prosumer ai products is about to explode.

[Alex Sharp @ajsharp]

@businessbarista @da_fant Narrator: all the context, was not, in fact, in the git repo

[Alex Sharp @ajsharp]

@IterIntellectus No it doesn’t. Youre either rage baiting or have no idea what youre talking about. Either way, this sucks.

[Alex Sharp @ajsharp]

@LizAnnSonders Great chart. The stuff you share is consistently unique and extremely poignant and revealing. Much appreciated, thank you.

[Alex Sharp @ajsharp]

Thank you for posting this. Wild how many supposed experts fundamentally misunderstand the relationship bw money supply and inflation. Nearly everyone I see thinks money supply go up, inflation go up. The real world is highly dynamic/complex, and in crisis, the financial system needs lots of dollars to properly function.

[Alex Sharp @ajsharp]

Quantitative evidence of Claude degradation.

My guess is Anthropic ran out of compute capacity sometime in feb due to their explosive growth since beginning of the year, and has some infra to dynamically scale thinking effort to load balance across the capacity they do have. Incredibly impressive they made to stay up as much as they do.

[ℏεsam @Hesamation]

AMD Senior AI Director confirms Claude has been nerfed. She analyzed Claude's session logs from Janurary to March:
> median thinking dropped from ~2,200 to ~600 chars
> API requests went up 80x from Feb to Mar. less thinking and failed attempts meaning more retries, burning more tokens, and spending more on tokens
> reads-per-edit dropped from 6.6x → 2.0x. model stops researching code before touching it.
> model tried to bail out or ask "should i continue" 173 times in 17 days (0 times before March 8).
> self-contradiction in reasoning ("oh wait, actually...") tripled.
> conventions like CLAUDE.md get ignored because there's less thinking budget to cross-check edits
> 5pm and 7pm PST are the worst hours, late night is significantly better. this means the thinking allocation is most likely GPU-load-sensitive.

[Alex Sharp @ajsharp]

@VivekVRao1 you probably need to determine what extreme is on a per-security basis, right? 5 min candles on some assets would prob trigger on things that would look like errors on another asset. thinking btc for example will often have insane deviations within a 5 min period

[Alex Sharp @ajsharp]

@joemccann @IBKR Totally. Have been doing this with options data, by far the best "free" option i've found, at least from a brokerage api

[Andrej Karpathy @karpathy]

Software horror: litellm PyPI supply chain attack.

Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords.

LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm.

Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks.

Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages.

Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

[Ben Hunt @EpsilonTheory]

While Trump dodged the draft, Robert Mueller volunteered for the Marines after graduating Princeton. He was awarded a Bronze Star with combat V, rescuing a wounded Marine while under fire. He was later shot in the thigh, awarded a Purple Heart, and returned to lead his platoon.