Released XAB so you can merge your slop with your bros' slop based on *intent* of the commits: https://x.com/angrymouse_dev/status/2039506565938422201
X (formerly Twitter)
mouse (@angrymouse_dev) on X
One of the most frustrating aspects of the development world in AI era is collaboration and specifically merging codebases that have been heavily modified by other people's agents and based on older version of upstream.
That's why I made XAB, AI based toolβ¦
That's why I made XAB, AI based toolβ¦
I think what made crypto space great and what it lacks right now is kind of the "startup" energy and people actually shipping things. Like back in the day, every coin had a product attached, often even blockchain, and now it's a rare occasion and "TECH play". Trenches need some source of liquidity to suck $$ from, and earlier there indeed were some actual *users* that you could suck $ off. Right now it's just PvPing each other without external liquidity source.
β€βπ₯3
I think once you have at least the ability to "vibe" code (actual ability to *code* is perfect tho), you can get much more ahead as every one of your launches can be "tech play". So I think right now what we lack is actually some tech-savvy person to kind of guide the trenches/devvs through to attract some external liq
β€βπ₯3π₯1π€1
If you don't bet on tech on tech that actually works and extracts *from users*, then you are basically betting on coins that extract from you and people alike you, so today you win tomorrow not, and it's not even good odds for you because of fees etc.
β€βπ₯1π1π₯1
Tried to use https://forgecode.dev/ instead of claude code/codex/etc for today, feels just... smarter. Apparently also performs better on benchmarks https://www.tbench.ai/leaderboard/terminal-bench/2.0
ForgeCode
ForgeCode is the world's #1 coding harness. Autonomously write, refactor, debug, and ship code β ranked #1 on TermBench 2.0 and built for the terminal.
β€βπ₯1β€1π₯1
A few actually realize, but the reason solana has so few tech projects is because of the cost to deploy actual smart contract to it. And no I don't mean to "deploy" a token on pump.fun or wherever - this just creates an *account* and doesn't cost nearly as much. When you actually want to upload smart contract *source* to have some custom logic, it costs about 8-10 sol just for that. You cannot even be sure you'll make that back as solana trends are very short lived, so this upfront investment makes deploying something actually technical, useful, etc quite unfeasible on solana, but it's a nice shitcoin factory IG. I think if you want to trade smarter, you just need to look on base/bsc as that's where real tech projects can launch (they won't pick solana because of things described above and eth is basically the same but also expensive for users, leaving only high performance, popular+cheap chains like base or BNB as options).
π2π€1
The Tech Quant
Tried to use https://forgecode.dev/ instead of claude code/codex/etc for today, feels just... smarter. Apparently also performs better on benchmarks https://www.tbench.ai/leaderboard/terminal-bench/2.0
Now that Claude Code has leaked and people are building their own "compatible" forks with other models, I wonder if we'll start seeing better harnesses that actually work with Claude Code subs without getting accounts termed (That arent through the Claude Code Agent SDK)
For reference, the standard Claude Code harness ranks 39th out of all tested harnesses π€
For reference, the standard Claude Code harness ranks 39th out of all tested harnesses π€
The Tech Quant
Now that Claude Code has leaked and people are building their own "compatible" forks with other models, I wonder if we'll start seeing better harnesses that actually work with Claude Code subs without getting accounts termed (That arent through the Claudeβ¦
Most likely the opposite, they will change some patterns and if you're using leaked version then just ban you
But tbh - I don't care if I get banned from anthropic if I get 20% improvement , fully worth it!
One interesting tendancy I notice right now is that the more code changes given person does in a timespan, the less he understands 'his' changes and overall codebase and further from reality he is