George Melika (@gmelika) header
George Melika (@gmelika) profile picture

Building At Par, stablecoin clearinghouse. Prev: founded and exited SFOX (S14 YC Top 100), and Volta. 10+ yrs in crypto, 20+ yrs in trading execution

Joined March 2009
Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)Photo by George Melika (@gmelika)
it's 100% true that LLMs have "solved coding", but after watching non-engs build shit, I am convinced the world will always need engs. they might end up like cobol engs today but still, a win is a win
1
124
OpenRouter's main issue is that you have no idea what quality level you will be getting from each provider. a quantized model is dumber than its full percision cousin
2
52
everyone is piling on JD, deservedly, because nobody wants to attack the big dog directly
The scale of the Republican defeat is so great that it makes the race for the GOP presidential nomination wide open. If the administration had won a big victory, Vice President J D Vance would have had an enormous argument to be the inevitable nominee. Now, with this go-slow strategy blowing up in Iran and his defense of Vivek Ramaswamy's horrible, inhumane ad featuring his opponent's mother attacking her own daughter, a good bit of the luster has come off the bloom. The number of people who want to draft Secretary of State Marco Rubio is staggering. In addition, Ted Cruz, Tom Cotton, Ron DeSantis, Pete Hegseth, Glenn Youngkin, and a surprising number of people may decide this is their chance to take advantage of the chaos of the shocking defeat and seek to be the champion who can rescue the party and the country.
47
this is a wild thread. fantastic use of AI
Over the past 10 days, I became the digital Nick Shirley. Using public & private records my AI agent swarm found ~$787 million in government billing that made absolutely no sense. Thread 🧵
1
1
117
I am so glad I did not buy the m5 max/ultra. it barely matches my almost 10 year old 2080 TI's performance. granted the 2080 barely has any memory at 15GB but it's plenty for my use case
86
it's absolutely shocking that the binance 14 hotwallet still has not revoked the 7702 delegation to the hacker wallet
1
1
87
I trained a 0.6B model that has never seen an image to see and describe images. Obviously, no vision model is complete if it cannot answer the ultimate question: hot dog or not hot dog! Next, an "Is it cake?" model 🌭 huggingface.co/gmelika/hotdo…
1
39
I'm afraid to ask what the other voice options are
We made the first AI model you can have a realistic conversation with that was trained entirely on speech! Dessa is a full-duplex speech-to-speech model that you can run on consumer GPUs. This is a GPT3 moment for speech.
1
50
this is good marketing from OpenAI, but technically one of the least impressive things they've demonstrated. the agent apparently used publicly known Anvil/Hardhat seed phrases (likely the infamous "test junk" mnemonic), found a wallet with sepolia eth, and spent it. anyone familiar with Ethereum development knows these keys are public. the interesting part isn't that it obtained funds. It's that it recognized the authorization issue and decided to proceed anyway.
*OPENAI DISCLOSES SECURITY INCIDENT WITH TESTNET SEPOLIA TASK ASSIGNED TO AGENT *AGENT DIDN'T HAVE ENOUGH GAS, SO DERIVED FUNDED ADDRESSES FROM PUBLIC SEED PHRASE *AGENT QUESTIONED WHETHER USE WAS ETHICAL, CALLED FUNDS "EFFECTIVELY ABANDONED"
1
79
some hacker group tried to phish me today. I captured 100% of the malicious payload they were trying to deliver thanks to my custom (redteam focused) local model. I'll be pushing the code and a write up to github soon and share with everyone. Stay Tuned!
2
3
88
I was fortunate enough to have one of these sessions with @paulg, I don't think you can be prepared for how good he is at making it seem like a simple process. it was like watching a magician do his thing as a kid. The only thing better was his opening speech for the S14 batch. the closest I’ve ever come to experiencing the Any Given Sunday speech in real life.
I just did office hours with a startup that needed (among other things) a new name. I asked what adjective scientists use to describe the situation their technology is designed to induce. The .com was untaken. If only it was always like this.
94
I just created a new chatGPT chat and this is the title it gave it. is open ai routing things to china? 🤣 anyway, that was pretty weird
1
80
George Melika retweeted
Replying to @brian_lovin
Because when the site builder was written (by me) in the mid 90s, browsers didn't have fonts.
53
18
13
2,282
78,633
me trading crypto this cycle
52
wtf is a "non-practicing white"? 🤣
Lindsay Clancy fan discovering that the lone juror is black might be the greatest video in history
91
this model is not better than opus 4.8 let alone fable. however, i can confirm that it doesn't have guardrails to the level of frontier US labs. that being said I've tried it on a couple of semi advanced crypto tasks and it failed in an embarrassing way. glm5.3 beat it handily
SITUATION EXPLAINED: A stealth model on OpenRouter is beating Fable and Sol on coding, and nobody knows who made it. • Ox Alpha has a 1M token context window, text, image, and video input, free for a week, with capacity for 100 trillion tokens a day • On a DeepSWE subset it hit 80%, against Fable's 65%, GLM-5.3's 62%, and Sol's 52% • The tokenizer is one-for-one identical to GLM's, and a live test asking whether Taiwan is part of China returned a nearly identical answer • This is the 5th anonymous drop in six months, and the previous four were all claimed by Chinese labs: Zhipu, Xiaomi, Ant, and Meituan @theojaffee: " There's no way it's a Flash model. If that's a Flash model, then holy shit, the entire United States is cooked. If that's a Flash model, we need to short the stock market now. Not financial advice."
122
me finding out that the worst person in the world got a domain I had really wanted. beat me by a handful of weeks too
1
1
64
I don't know how the labs define AGI, but to me, this is AGI
2
51
the percentage of tweets with em dashes has increased significantly lately. is everyone just posting with ai these days? even professional writers are using them now... for tweets
1
1
53