Загрузка видео...

Не удалось загрузить видео

На главную

Your JEV agent knows the answer. It doesn't know whether it should act on it every call returns two things: the answer, and the confidence behind it. most setups only read the first one. confidence comes from the spread. same 4 options, one distribution concentrated at 0.91, one flat...

0x_rody's profile picture

rody

4,916 subscribers

30,276 просмотров • 1 месяц назад •via X (Twitter)

Наука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 14

Фото профиля PolyBender
PolyBender1 месяц назад

jev always knows the right answer

Фото профиля rody
rody1 месяц назад

much appreciated!

Фото профиля Amir
Amir1 месяц назад

that point about confidence versus permission hit home because we assumed a high score meant the job actually saved

Фото профиля Moysha
Moysha1 месяц назад

booked this

Фото профиля rody
rody1 месяц назад

thank you fam!

Фото профиля Hussain Hashim | Building SundayBack
Hussain Hashim | Building SundayBack1 месяц назад

@0x_rody missed opportunities in planning for that spread. not enough people analyze it post-call.

Фото профиля cloutusf
cloutusf1 месяц назад

0.91 confidence that it's done, and the file's still sitting there unsaved. Somehow both show up on the same call.

Фото профиля Brian Hadu
Brian Hadu1 месяц назад

how do you see teams effectively using the confidence metric in practice?

Фото профиля magsimich
magsimich1 месяц назад

The confidence part is easy to overlook

Фото профиля Fajar M Reza
Fajar M Reza1 месяц назад

Confidence estimates make agent actions easier to govern than answers alone.

Фото профиля Len
Len1 месяц назад

Fake

Фото профиля Sharapov Andrei
Sharapov Andrei1 месяц назад

Pinning the model version isn't enough if the choice set changes. Confidence is computed from the spread over K options — add a hedge choice and last week's 0.91 is no longer the same refund bar.

Фото профиля BonBonaz
BonBonaz1 месяц назад

my confidence comes from hopium not spread

Фото профиля heath
heath1 месяц назад

the confident “done” without checking the file is terrifying lol. I built helix to manage my AI agents, and this is exactly the kind of action boundary I want visible.

Похожие видео

LLMs vs. Jev, clearly explained! LLMs are great, and the ceiling is one you can watch scroll past: an LLM writes the answer one token at a time. give it a failed deploy and four decisions, and it produces a small JSON object where every token depends on the one before it. token nine cannot exist until token eight does, so four decisions that had nothing to do with each other just stood in a queue. then your code parses it, validates the shape, and retries when the shape is wrong. Jev fixes this without being a smaller or faster model: it removes the order. one turn on that deploy has to know: → whether the incident is urgent → which team owns it → whether the next command is risky → whether the task is actually done you declare the questions and the answer type upfront, and all four come back together, typed, with a probability on each. three primitives cover almost every fork in an agent: 1. **Choice** picks one of up to 255 options you define, like engineering, billing or sales. 2. **Score** places the state on an ordered scale you define, like low, medium or high risk. 3. **Noul** returns the probability that a yes-or-no condition is true. here is the sentence that resolves the whole confusion: text is a line you have to walk. an answer space is a room you see all of at once. ↳ generation: one order you cannot change, one string at the end, a shape you hope holds ↳ evaluation: no order at all, typed answers, a probability on every option Prompts → Agents → Loops → Graphs → Jev the probabilities matter more than the answer. ↳ engineering at 0.91 against billing at 0.09 is a route you can automate ↳ 0.52 against 0.46 is a coin flip wearing a label, and the label alone never told you which one you got that last one catches careful people. an LLM would have said "engineering" in a confident sentence and given you no way to know the race was that close. thresholds live in your code, one per action, scaled to what being wrong costs. it works when the options are known and the call depends on meaning. it is not for writing, code, arithmetic, or anything where question two needs the answer to question one. and the one that eats whole nights: type safety prevents malformed output, not incorrect judgment. Jev cannot return an option outside your schema, and it can still pick the wrong valid one with confidence. a schema-valid mistake refunds the wrong customer just as fast. an LLM writes new language when the answer space is open. Jev evaluates known paths when the answer space is closed. below i have quoted my full breakdown on Jev. it covers the three primitives, the parallel battery, the thresholds, and where it does not belong. save this and read it below ↓
0:35

LLMs vs. Jev, clearly explained! LLMs are great, and the ceiling is one you can watch scroll past: an LLM writes the answer one token at a time. give it a failed deploy and four decisions, and it produces a small JSON object where every token depends on the one before it. token nine cannot exist until token eight does, so four decisions that had nothing to do with each other just stood in a queue. then your code parses it, validates the shape, and retries when the shape is wrong. Jev fixes this without being a smaller or faster model: it removes the order. one turn on that deploy has to know: → whether the incident is urgent → which team owns it → whether the next command is risky → whether the task is actually done you declare the questions and the answer type upfront, and all four come back together, typed, with a probability on each. three primitives cover almost every fork in an agent: 1. **Choice** picks one of up to 255 options you define, like engineering, billing or sales. 2. **Score** places the state on an ordered scale you define, like low, medium or high risk. 3. **Noul** returns the probability that a yes-or-no condition is true. here is the sentence that resolves the whole confusion: text is a line you have to walk. an answer space is a room you see all of at once. ↳ generation: one order you cannot change, one string at the end, a shape you hope holds ↳ evaluation: no order at all, typed answers, a probability on every option Prompts → Agents → Loops → Graphs → Jev the probabilities matter more than the answer. ↳ engineering at 0.91 against billing at 0.09 is a route you can automate ↳ 0.52 against 0.46 is a coin flip wearing a label, and the label alone never told you which one you got that last one catches careful people. an LLM would have said "engineering" in a confident sentence and given you no way to know the race was that close. thresholds live in your code, one per action, scaled to what being wrong costs. it works when the options are known and the call depends on meaning. it is not for writing, code, arithmetic, or anything where question two needs the answer to question one. and the one that eats whole nights: type safety prevents malformed output, not incorrect judgment. Jev cannot return an option outside your schema, and it can still pick the wrong valid one with confidence. a schema-valid mistake refunds the wrong customer just as fast. an LLM writes new language when the answer space is open. Jev evaluates known paths when the answer space is closed. below i have quoted my full breakdown on Jev. it covers the three primitives, the parallel battery, the thresholds, and where it does not belong. save this and read it below ↓

Hanako

44,238 просмотров • 1 месяц назад

this is pure f*cking gold for anyone running coding agents Jev founder, Diogo Almeida, wrote a guide on building a Jev harness the expensive model only writes the code. every fork around it goes to Jev the bill: > 6,000 forks a day on a frontier model: $180 > the same 6,000 forks on Jev: $0.25 > the whole harness, patches included: about $24 99 percent of what's left is the code itself, which is the only part you should be paying for what's inside: > the turn snapshot, five hooks, the cascade > thresholds read off your own receipts > the one rule a model never gets a vote on one evening of setup and your agent stops paying frontier prices to answer yes or no ↓
0:30

this is pure f*cking gold for anyone running coding agents Jev founder, Diogo Almeida, wrote a guide on building a Jev harness the expensive model only writes the code. every fork around it goes to Jev the bill: > 6,000 forks a day on a frontier model: $180 > the same 6,000 forks on Jev: $0.25 > the whole harness, patches included: about $24 99 percent of what's left is the code itself, which is the only part you should be paying for what's inside: > the turn snapshot, five hooks, the cascade > thresholds read off your own receipts > the one rule a model never gets a vote on one evening of setup and your agent stops paying frontier prices to answer yes or no ↓

Hanako

20,837 просмотров • 1 месяц назад

Here is the full walkthrough if you want to watch it work. A real support ticket goes in, and Jev comes back with the team that owns it, a frustration score, and a probability that it reads as urgent. The confidence value on each answer decides what gets handled automatically and what a person sees.
10:25

Here is the full walkthrough if you want to watch it work. A real support ticket goes in, and Jev comes back with the team that owns it, a frustration score, and a probability that it reads as urgent. The confidence value on each answer decides what gets handled automatically and what a person sees.

Netlify

10,336 просмотров • 1 месяц назад

ANTHROPIC LEAKED A $4.2M FILE WHERE 4 AGENTS BUILD THE JEV SYSTEM IN 60 MINUTES, DOING $100K A MONTH four agents cost $0, and the system they assemble in an hour generates $100K a month scout → builder → critic → operator → JEV every agent is pinned to one exact model version, and none of them runs on "latest" the scout on Haiku 4.5 reads the market, the builder on Sonnet 5 builds everything, the critic on Opus 5 decides what ships the operator runs Haiku through the batch API at half price - the one agent that never has to hurry the builder takes 59% of the whole hour, because it's the one writing, and writing costs 5x more than reading short answers are cheap answers - and the whole economics rests on that one rule pin the versions in ROOT.md and next month JEV behaves exactly the same point it at "latest" and the model underneath updates quietly, no error, and the answers start to drift $100K a month is 334 customers at $299 four versions, one system - pin them and JEV stays JEV save this and paste it into Claude Code - let it build you these 4 agents ↓
1:28

ANTHROPIC LEAKED A $4.2M FILE WHERE 4 AGENTS BUILD THE JEV SYSTEM IN 60 MINUTES, DOING $100K A MONTH four agents cost $0, and the system they assemble in an hour generates $100K a month scout → builder → critic → operator → JEV every agent is pinned to one exact model version, and none of them runs on "latest" the scout on Haiku 4.5 reads the market, the builder on Sonnet 5 builds everything, the critic on Opus 5 decides what ships the operator runs Haiku through the batch API at half price - the one agent that never has to hurry the builder takes 59% of the whole hour, because it's the one writing, and writing costs 5x more than reading short answers are cheap answers - and the whole economics rests on that one rule pin the versions in ROOT.md and next month JEV behaves exactly the same point it at "latest" and the model underneath updates quietly, no error, and the answers start to drift $100K a month is 334 customers at $299 four versions, one system - pin them and JEV stays JEV save this and paste it into Claude Code - let it build you these 4 agents ↓

Sprytix

38,104 просмотров • 1 месяц назад

call me crazy but company knowledge should look more like this. > every file becomes a node > every import, call, and dependency becomes an edge > every answer points back to the exact code behind it click one class and the system shows what it depends on, what depends on it, and where it sits inside the architecture. now apply the same idea to the company: decisions. projects. people. policies. workflows. learnings. AI should not search through a pile of documents and guess how they connect. it should navigate the company as a system. don't bookmark this if it crosses your timeline. paste the full article into your agent and tell it to build the first version for your company.
0:30

call me crazy but company knowledge should look more like this. > every file becomes a node > every import, call, and dependency becomes an edge > every answer points back to the exact code behind it click one class and the system shows what it depends on, what depends on it, and where it sits inside the architecture. now apply the same idea to the company: decisions. projects. people. policies. workflows. learnings. AI should not search through a pile of documents and guess how they connect. it should navigate the company as a system. don't bookmark this if it crosses your timeline. paste the full article into your agent and tell it to build the first version for your company.

J.B.

57,762 просмотров • 2 месяцев назад

ANTHROPIC LEAKED A FILE THEIR LEAD ENGINEER SPENT 4 YEARS BUILDING AT A COST OF $3.2M - ONE SUPER AGENT WITH JEV HOLDS A BUSINESS DOING $2.4M A YEAR the agent costs $0, and in month one it's already pulling $100K+ read → decide → build → ship → watch → back to reading 24 cycles a day, and each one starts exactly where the last one stopped it doesn't sleep, doesn't forget and never asks what to do next in one night it runs more cycles than a team closes in a quarter JEV holds the context, the budget and the stop condition between hours - without it every cycle would start from zero it picks one change an hour, not twenty - which is exactly why nothing breaks if things get worse it rolls back on its own, no meeting and no explanation the price of the loop doesn't grow with the revenue: it's the same at $0 and at $200,000 a month $2.4M a year is 669 customers, and that's only true while all 669 keep paying every month one agent is one point of failure: the night the loop stopped, nobody noticed until morning it ships the changes, it doesn't find the customers - that part is still yours save this and paste it into Claude Code - let it build you this loop ↓
0:59

ANTHROPIC LEAKED A FILE THEIR LEAD ENGINEER SPENT 4 YEARS BUILDING AT A COST OF $3.2M - ONE SUPER AGENT WITH JEV HOLDS A BUSINESS DOING $2.4M A YEAR the agent costs $0, and in month one it's already pulling $100K+ read → decide → build → ship → watch → back to reading 24 cycles a day, and each one starts exactly where the last one stopped it doesn't sleep, doesn't forget and never asks what to do next in one night it runs more cycles than a team closes in a quarter JEV holds the context, the budget and the stop condition between hours - without it every cycle would start from zero it picks one change an hour, not twenty - which is exactly why nothing breaks if things get worse it rolls back on its own, no meeting and no explanation the price of the loop doesn't grow with the revenue: it's the same at $0 and at $200,000 a month $2.4M a year is 669 customers, and that's only true while all 669 keep paying every month one agent is one point of failure: the night the loop stopped, nobody noticed until morning it ships the changes, it doesn't find the customers - that part is still yours save this and paste it into Claude Code - let it build you this loop ↓

Sprytix

10,367 просмотров • 1 месяц назад

David Jerison, the mathematician who spent decades teaching the exact method for finding the optimal choice when the odds are stacked against you: "I used to think the best decision was whichever option looked strongest at first glance. Then I proved that the real answer almost always hides at a point nobody would guess just by looking, and checking only the obvious choices gets you the worst possible outcome, not the best." this is the exact method quant desks lean on to find the one allocation that survives every constraint thrown at it, and it's been sitting free in a public MIT lecture for almost twenty years. strip away the notation and the mechanism is simple. every optimization problem has a handful of candidate points where the best or worst answer could be hiding, and the obvious middle-of-the-road guess is almost never one of them. check only the points that feel natural, and you don't just miss the best answer. you can land on the exact opposite, the worst possible one, without ever realizing it. nobody presenting a "risk-optimized" portfolio out loud admits how easy it is to stop checking one step too early. zoom out to how this plays out sizing a position or allocating risk under real constraints today. the instinct is to test the option that feels balanced and call it done, when the actual edge is almost always sitting at an extreme nobody thought to check. the industry sells a clean, confident number as proof an allocation is optimal. but that number means nothing until every boundary has been checked, because the same method that finds the best case can just as easily hand you the worst one in disguise. the right answer was never the one that looked most reasonable. it was the one nobody bothered to check.
49:15

David Jerison, the mathematician who spent decades teaching the exact method for finding the optimal choice when the odds are stacked against you: "I used to think the best decision was whichever option looked strongest at first glance. Then I proved that the real answer almost always hides at a point nobody would guess just by looking, and checking only the obvious choices gets you the worst possible outcome, not the best." this is the exact method quant desks lean on to find the one allocation that survives every constraint thrown at it, and it's been sitting free in a public MIT lecture for almost twenty years. strip away the notation and the mechanism is simple. every optimization problem has a handful of candidate points where the best or worst answer could be hiding, and the obvious middle-of-the-road guess is almost never one of them. check only the points that feel natural, and you don't just miss the best answer. you can land on the exact opposite, the worst possible one, without ever realizing it. nobody presenting a "risk-optimized" portfolio out loud admits how easy it is to stop checking one step too early. zoom out to how this plays out sizing a position or allocating risk under real constraints today. the instinct is to test the option that feels balanced and call it done, when the actual edge is almost always sitting at an extreme nobody thought to check. the industry sells a clean, confident number as proof an allocation is optimal. but that number means nothing until every boundary has been checked, because the same method that finds the best case can just as easily hand you the worst one in disguise. the right answer was never the one that looked most reasonable. it was the one nobody bothered to check.

MindArch

16,232 просмотров • 2 месяцев назад

How does Jev work? Jev takes two things: your input and an output schema. The schema gives the structure of the answer you want. I send an email. My schema asks for two fields: is spam, and a category from marketing, finance, or spam. Jev returns: - is spam: 0.90 - category: spam That is a 90% chance of spam. Jev selects the category only from the list I give it. It writes no sentences. The 0.90 is a number object, so my code can use it immediately without extra processing. Jev shows no reasoning. You get the answer and nothing more, so some people do not trust it. Use Jev for choices your code must make fast and at low cost: spam or normal, buy or no buy, reply or escalate.
8:26

How does Jev work? Jev takes two things: your input and an output schema. The schema gives the structure of the answer you want. I send an email. My schema asks for two fields: is spam, and a category from marketing, finance, or spam. Jev returns: - is spam: 0.90 - category: spam That is a 90% chance of spam. Jev selects the category only from the list I give it. It writes no sentences. The 0.90 is a number object, so my code can use it immediately without extra processing. Jev shows no reasoning. You get the answer and nothing more, so some people do not trust it. Use Jev for choices your code must make fast and at low cost: spam or normal, buy or no buy, reply or escalate.

The Startup Ideas Podcast (SIP) 🧃

12,032 просмотров • 1 месяц назад

this is crazy... the guy reverse-engineered his Meta Ray-Bans 3 with Opus 5.5 into a 3,000+ Elo chess engine he built AI chess cheat glasses on Opus 5.5 + Jev and open-sourced the whole workflow here's how it works step by step: step 1 → install it on your Mac with one command: npx chesshacker setup. It downloads Stockfish from GitHub and verifies the checksum step 2 → put on Meta Ray-Bans and video-call your own Mac on WhatsApp. The glasses see the board, the Mac sees what the glasses see step 3 → the Mac grabs frames straight from the call window. About 1 per second while it waits for the opponent, 1 every 3 seconds on your turn step 4 → Jev runs a cheap pre-check on every frame: is the board readable, is a hand over it, did a piece move? No change means no expensive AI call step 5 → only then Opus 5.5 reads the board. The app compares it with the position it already knows and accepts a move only if it's legal step 6 → Stockfish thinks for half a second and picks the best reply, with an eval like +1.5 or "mate in 3" step 7 → macOS speaks it through a virtual mic into the WhatsApp call: "Rook a to d1". You hear it in your ear step 8 → on the Mac a command center shows everything live: whose turn it is, the suggested move with a confidence bar, the move list and a log under the hood: → if Opus loses track of the board, the glasses say "Lost track of the board" and you fix it with undo or by typing the move → if Stockfish crashes, it restarts once and says "Something went wrong. Still trying" → every game is saved as a PGN file, API keys live in the macOS Keychain → an overlay draws the suggested move as an arrow on the live video, ready for OBS Clone his workflow below Then read the guide on how to reverse-engineer anything with Opus 5.5.
2:15

this is crazy... the guy reverse-engineered his Meta Ray-Bans 3 with Opus 5.5 into a 3,000+ Elo chess engine he built AI chess cheat glasses on Opus 5.5 + Jev and open-sourced the whole workflow here's how it works step by step: step 1 → install it on your Mac with one command: npx chesshacker setup. It downloads Stockfish from GitHub and verifies the checksum step 2 → put on Meta Ray-Bans and video-call your own Mac on WhatsApp. The glasses see the board, the Mac sees what the glasses see step 3 → the Mac grabs frames straight from the call window. About 1 per second while it waits for the opponent, 1 every 3 seconds on your turn step 4 → Jev runs a cheap pre-check on every frame: is the board readable, is a hand over it, did a piece move? No change means no expensive AI call step 5 → only then Opus 5.5 reads the board. The app compares it with the position it already knows and accepts a move only if it's legal step 6 → Stockfish thinks for half a second and picks the best reply, with an eval like +1.5 or "mate in 3" step 7 → macOS speaks it through a virtual mic into the WhatsApp call: "Rook a to d1". You hear it in your ear step 8 → on the Mac a command center shows everything live: whose turn it is, the suggested move with a confidence bar, the move list and a log under the hood: → if Opus loses track of the board, the glasses say "Lost track of the board" and you fix it with undo or by typing the move → if Stockfish crashes, it restarts once and says "Something went wrong. Still trying" → every game is saved as a PGN file, API keys live in the macOS Keychain → an overlay draws the suggested move as an arrow on the live video, ready for OBS Clone his workflow below Then read the guide on how to reverse-engineer anything with Opus 5.5.

Movez

398,388 просмотров • 7 дней назад

Early on in my career, I had one business and I was still figuring it out. The decisions in front of me felt like everything was on the line. I needed a model. Something that could make every call clearer, big or small. What I figured out: all decision-making is value clarification. It comes down to what’s most important to you. Two people can look at the same situation and make completely different calls. Both right. Because the answer always comes back to what you value most. Next time a decision doesn’t feel clear, run it through OOC-EMR.
2:14

Early on in my career, I had one business and I was still figuring it out. The decisions in front of me felt like everything was on the line. I needed a model. Something that could make every call clearer, big or small. What I figured out: all decision-making is value clarification. It comes down to what’s most important to you. Two people can look at the same situation and make completely different calls. Both right. Because the answer always comes back to what you value most. Next time a decision doesn’t feel clear, run it through OOC-EMR.

Tony Robbins

62,770 просмотров • 5 месяцев назад

one agent → a loop → a graph → a system that throttles itself Google just put the whole ladder in one free course. most people never leave the first rung. 2 hours 4 minutes, taught by two Google Cloud engineers. no signup, no paywall. 17:55 - build your first agent 39:55 - loop engineering: iterate, check, break 1:13:25 - graph engineering 1:35:27 - agents that throttle themselves 1:56:19 - the full graph for multi-agent systems the fourth one is the step nobody talks about. an agent that cannot throttle itself is not autonomous, it is just unattended. everyone builds one agent and calls it done. this is the version where the agents wire themselves into a graph and decide their own rate. and the thing that decides when one of them is allowed to stop is not in the course. that part you build yourself. bookmark and watch it today, then read the eval setup below
2:04:19

one agent → a loop → a graph → a system that throttles itself Google just put the whole ladder in one free course. most people never leave the first rung. 2 hours 4 minutes, taught by two Google Cloud engineers. no signup, no paywall. 17:55 - build your first agent 39:55 - loop engineering: iterate, check, break 1:13:25 - graph engineering 1:35:27 - agents that throttle themselves 1:56:19 - the full graph for multi-agent systems the fourth one is the step nobody talks about. an agent that cannot throttle itself is not autonomous, it is just unattended. everyone builds one agent and calls it done. this is the version where the agents wire themselves into a graph and decide their own rate. and the thing that decides when one of them is allowed to stop is not in the course. that part you build yourself. bookmark and watch it today, then read the eval setup below

Hanako

53,692 просмотров • 3 месяцев назад

your agent reviewing its own work is not a check. it is a second opinion from the same source. this is the most common gap in agent systems and it hides in plain sight, because the step exists. there is a review. it just cannot do the thing you think it does. here is the mechanism. the model produced an output from a context. you then ask the same model, holding the same context, whether that output is correct. it answers fluently, because that is what it does. and the answer is drawn from the same distribution that produced the thing being judged. same weights, same window, same blind spots. if the reason the output is wrong is something the model does not know, the review does not know it either. if the reason is something the context does not contain, the review has the same context. the failure mode and the detector share a cause. > why it feels like it works because most of the time the output is fine, and the review says fine. agreement is not evidence of detection. a reviewer that says pass on everything agrees with reality most of the time too. what you actually want to measure is what happens on the cases that are wrong. that is the only place a check earns its name, and it is exactly the place where a self-review is weakest. there is research on this. Huang and colleagues at DeepMind showed at ICLR 2024 that intrinsic self-correction, revising without external grounding, does not reliably help and often makes things worse. > what to actually do move the check outside the model. a test that runs, a schema that validates, a file that exists or does not, an exit code from something you did not write. these are not smarter than the model. they are just not correlated with it, and that is the entire value. when the judgement genuinely needs a model, at minimum use a different family. same family means shared blind spots, and frontier judges measurably inflate scores for outputs that look like their own. and split the work by kind. anything objectively checkable goes to code. only the genuinely semantic calls go to a judge, and those get a rubric written as one line. a review inside the loop tells you the model is confident. a check outside it tells you whether the work is done. save this - then read the eval setup below
0:50

your agent reviewing its own work is not a check. it is a second opinion from the same source. this is the most common gap in agent systems and it hides in plain sight, because the step exists. there is a review. it just cannot do the thing you think it does. here is the mechanism. the model produced an output from a context. you then ask the same model, holding the same context, whether that output is correct. it answers fluently, because that is what it does. and the answer is drawn from the same distribution that produced the thing being judged. same weights, same window, same blind spots. if the reason the output is wrong is something the model does not know, the review does not know it either. if the reason is something the context does not contain, the review has the same context. the failure mode and the detector share a cause. > why it feels like it works because most of the time the output is fine, and the review says fine. agreement is not evidence of detection. a reviewer that says pass on everything agrees with reality most of the time too. what you actually want to measure is what happens on the cases that are wrong. that is the only place a check earns its name, and it is exactly the place where a self-review is weakest. there is research on this. Huang and colleagues at DeepMind showed at ICLR 2024 that intrinsic self-correction, revising without external grounding, does not reliably help and often makes things worse. > what to actually do move the check outside the model. a test that runs, a schema that validates, a file that exists or does not, an exit code from something you did not write. these are not smarter than the model. they are just not correlated with it, and that is the entire value. when the judgement genuinely needs a model, at minimum use a different family. same family means shared blind spots, and frontier judges measurably inflate scores for outputs that look like their own. and split the work by kind. anything objectively checkable goes to code. only the genuinely semantic calls go to a judge, and those get a rubric written as one line. a review inside the loop tells you the model is confident. a check outside it tells you whether the work is done. save this - then read the eval setup below

Hanako

14,325 просмотров • 2 месяцев назад

31 guides from the team that built Grok Bot, and 99% of people haven't opened one > 15 official guides on x ai website > 16 articles from the engineers and PMs behind the product there is one for marketing, SDRs, recruiting, legal ops, support, engineering and more -> pick the one that matches your job -> follow it -> ship your bot most people are still guessing how to set it up the answer is already written, by the team that shipped it save this and read one tonight all 31 links below ⭣
0:22

31 guides from the team that built Grok Bot, and 99% of people haven't opened one > 15 official guides on x ai website > 16 articles from the engineers and PMs behind the product there is one for marketing, SDRs, recruiting, legal ops, support, engineering and more -> pick the one that matches your job -> follow it -> ship your bot most people are still guessing how to set it up the answer is already written, by the team that shipped it save this and read one tonight all 31 links below ⭣

unicode

10,899 просмотров • 1 месяц назад

This is f*cking insane. US researchers published "The End of the Judgment Call" and the whole paper is one split: Claude builds, Jev decides. The case is simple. An overnight Claude loop asks itself around 600 yes-or-no questions. Safe to run. Next step. Goal met. Each one costs about 4 cents on Fable 5.1 and takes seconds. Jev came out of stealth on September 15 from a co-author of InstructGPT. It answers typed questions only, with a calibrated probability, in about 100 milliseconds, at $0.042 per million input tokens. When it says 0.9 it is right nine times in ten. That is a number your code can act on at 3am with nobody watching. Three hooks in Claude Code and the judgment calls are gone: a 100 ms gate on every bash command, a stop hook that knows when the work is really done, a cron that only wakes Claude for reproducible bugs. The bill for 30 nights goes from $765 to $3.02. What Claude spends on building does not change. The article below has all three builds with code. The entire paper in one line: the cheapest Claude turn is the one that never needed to start ↓
0:31

This is f*cking insane. US researchers published "The End of the Judgment Call" and the whole paper is one split: Claude builds, Jev decides. The case is simple. An overnight Claude loop asks itself around 600 yes-or-no questions. Safe to run. Next step. Goal met. Each one costs about 4 cents on Fable 5.1 and takes seconds. Jev came out of stealth on September 15 from a co-author of InstructGPT. It answers typed questions only, with a calibrated probability, in about 100 milliseconds, at $0.042 per million input tokens. When it says 0.9 it is right nine times in ten. That is a number your code can act on at 3am with nobody watching. Three hooks in Claude Code and the judgment calls are gone: a 100 ms gate on every bash command, a stop hook that knows when the work is really done, a cron that only wakes Claude for reproducible bugs. The bill for 30 nights goes from $765 to $3.02. What Claude spends on building does not change. The article below has all three builds with code. The entire paper in one line: the cheapest Claude turn is the one that never needed to start ↓

slash1s

14,231 просмотров • 1 месяц назад

stop scrolling and clone this before it gets paywalled. someone put their entire trading loop on GitHub. the model, the order logic, the dashboard. all of it. it makes one decision every ~300 milliseconds and quotes a fresh limit order on every single block. the brain is Jev. it can't write a word. it looks at the order book and returns buy or sell with a probability. what the loop does each block: -> reads the live order book -> asks Jev: buy or sell? -> posts a post-only limit order one tick inside the touch -> replaces the last order -> waits for a taker to hit it the useful part: it never crosses the spread. fills only happen when someone else takes the order, so the bot is built to earn the spread instead of paying it. and every block is logged: the decision, the probabilities, the latency, the quote. the setup: git clone cp .env.example .env bun install bun run start -> with no private key it dry-runs: real book, real decisions, simulated fills then in .env: MODEL=jev TYPESAFE_AI_API_KEY=your-key -> swaps the built-in momentum stand-in for Jev want the full desk instead of one loop? -> QuantDinger ships Jev as a pre-trade gate: research, Python strategies, backtest, paper and live the workflow: order book in -> Jev decides -> code prices the order -> the market fills it or doesn't start here: one clone -> dry run -> watch 1,000 blocks of decisions before a single real order the catch: fast is not the same as right. quoting inside the spread can still lose money when the price moves against you. none of this is an ATM. > let the big model do the research. > let Jev make the call. > let code place the order. clone it now. dry-run it tonight. decide tomorrow.
0:30

stop scrolling and clone this before it gets paywalled. someone put their entire trading loop on GitHub. the model, the order logic, the dashboard. all of it. it makes one decision every ~300 milliseconds and quotes a fresh limit order on every single block. the brain is Jev. it can't write a word. it looks at the order book and returns buy or sell with a probability. what the loop does each block: -> reads the live order book -> asks Jev: buy or sell? -> posts a post-only limit order one tick inside the touch -> replaces the last order -> waits for a taker to hit it the useful part: it never crosses the spread. fills only happen when someone else takes the order, so the bot is built to earn the spread instead of paying it. and every block is logged: the decision, the probabilities, the latency, the quote. the setup: git clone cp .env.example .env bun install bun run start -> with no private key it dry-runs: real book, real decisions, simulated fills then in .env: MODEL=jev TYPESAFE_AI_API_KEY=your-key -> swaps the built-in momentum stand-in for Jev want the full desk instead of one loop? -> QuantDinger ships Jev as a pre-trade gate: research, Python strategies, backtest, paper and live the workflow: order book in -> Jev decides -> code prices the order -> the market fills it or doesn't start here: one clone -> dry run -> watch 1,000 blocks of decisions before a single real order the catch: fast is not the same as right. quoting inside the spread can still lose money when the price moves against you. none of this is an ATM. > let the big model do the research. > let Jev make the call. > let code place the order. clone it now. dry-run it tonight. decide tomorrow.

kiosa

21,466 просмотров • 21 дней назад

They taught you one chapter and quietly skipped the rest. Stop scrolling, because this is the story that goes silent the moment someone asks about it. Millions of people, a desert that swallowed the evidence, and one country that didn’t make it a crime until 2007. Every fact makes the next one harder to believe, and the question at the end is the one nobody wants to answer on camera. One detail in the final seconds changes how you see all of it. Watch to the last second, then send it to someone who thinks they already know the story. WATCH FULL EPISODE BELOW!!!
2:44

They taught you one chapter and quietly skipped the rest. Stop scrolling, because this is the story that goes silent the moment someone asks about it. Millions of people, a desert that swallowed the evidence, and one country that didn’t make it a crime until 2007. Every fact makes the next one harder to believe, and the question at the end is the one nobody wants to answer on camera. One detail in the final seconds changes how you see all of it. Watch to the last second, then send it to someone who thinks they already know the story. WATCH FULL EPISODE BELOW!!!

Jillian Michaels

45,789 просмотров • 8 дней назад

Agents vs. Graphs, clearly explained! spawning more agents is great, but it has a ceiling nobody says out loud: five agents is a count. a graph is a shape. only one of them changes the answer. point five agents at the same pile with the same window and they converge. the first one writes a finding, the rest read it, and all five reports centre on the same thing. you paid five times for one opinion with four echoes. Graph engineering fixes this by moving the decision up a layer: not how many agents, but who is allowed to look at what. you need both. here's how it works: ↳ the count buys you throughput. five things happening instead of one ↳ the shape buys you coverage. five different things happening instead of the same one five times Prompts → Context → Harness → Agents → Graphs the node that does this is the splitter, and it decides more than any other node in the system. cut a repository by folder and four workers audit the same three files. cut it by blast radius and each one sees something the others cannot. the trick is being selective about what each lane is allowed to see. separate contexts are not a nice-to-have, they are the mechanism. if two agents are meant to produce different things, they must not share a window. if they are meant to produce the same thing, you did not need two agents. one thing to know before you scale it. a branch that throws does not reject the batch. it resolves to null, and that is the containment. which means your merge quietly receives a short list. ↳ filter the nulls before the merge, or one dead lane poisons the whole result ↳ never index a merge by position. eight good branches and one failure will shift everything by one, silently skip that and the run looks like it worked. the output is just missing a lane, and nothing errored. and the one that eats whole nights: multi-agent setups can use up to fifteen times the total tokens of a single chat, because every lane reloads its own core. you are trading total tokens for a clean main window. usually the right trade, always a choice. below i have quoted my full guide on graph engineering. it covers the three topologies, the verifier patterns, and where the gate should actually open. save this and read it below ↓
0:49

Agents vs. Graphs, clearly explained! spawning more agents is great, but it has a ceiling nobody says out loud: five agents is a count. a graph is a shape. only one of them changes the answer. point five agents at the same pile with the same window and they converge. the first one writes a finding, the rest read it, and all five reports centre on the same thing. you paid five times for one opinion with four echoes. Graph engineering fixes this by moving the decision up a layer: not how many agents, but who is allowed to look at what. you need both. here's how it works: ↳ the count buys you throughput. five things happening instead of one ↳ the shape buys you coverage. five different things happening instead of the same one five times Prompts → Context → Harness → Agents → Graphs the node that does this is the splitter, and it decides more than any other node in the system. cut a repository by folder and four workers audit the same three files. cut it by blast radius and each one sees something the others cannot. the trick is being selective about what each lane is allowed to see. separate contexts are not a nice-to-have, they are the mechanism. if two agents are meant to produce different things, they must not share a window. if they are meant to produce the same thing, you did not need two agents. one thing to know before you scale it. a branch that throws does not reject the batch. it resolves to null, and that is the containment. which means your merge quietly receives a short list. ↳ filter the nulls before the merge, or one dead lane poisons the whole result ↳ never index a merge by position. eight good branches and one failure will shift everything by one, silently skip that and the run looks like it worked. the output is just missing a lane, and nothing errored. and the one that eats whole nights: multi-agent setups can use up to fifteen times the total tokens of a single chat, because every lane reloads its own core. you are trading total tokens for a clean main window. usually the right trade, always a choice. below i have quoted my full guide on graph engineering. it covers the three topologies, the verifier patterns, and where the gate should actually open. save this and read it below ↓

Hanako

97,154 просмотров • 2 месяцев назад

I just built a Jev X Viral Post Analyser. 100,000 viral X posts. 20.4 seconds. $0.67. Claude Opus 5, same corpus, same clock, got through 214 posts and spent $0.98. per post that is ~680x cheaper the full Opus pass would have run $458. viral analysis is the perfect Jev job. • it is not writing, it is 14 yes/no calls per post: > does the hook open a loop, > is there a number in the first line, > is the proof real or claimed. classification, not prose. • what it found: 1,220 posts broke into the top 1%. baseline 1.22%. > superlative claim - 2.34% viral. 1.92x baseline > contrarian take - 1.59%. 1.31x > launch / tool drop - 1.46%. 1.19x and numbered lists, the thing everyone writes: 0.55%. below baseline. the most used hook is the least viral one. full stop. • what you are watching: left is the post under analysis, right is Jev answering 14 typed questions about it, each with a confidence score. the run stops at 20.4s because that is when Jev finished all 100k. pulled the corpus through a few X APIs, one parallel pass into Jev. should I drop it to public? Read my latest article on Jev Engineering below and turn your ideas into reality.
0:25

I just built a Jev X Viral Post Analyser. 100,000 viral X posts. 20.4 seconds. $0.67. Claude Opus 5, same corpus, same clock, got through 214 posts and spent $0.98. per post that is ~680x cheaper the full Opus pass would have run $458. viral analysis is the perfect Jev job. • it is not writing, it is 14 yes/no calls per post: > does the hook open a loop, > is there a number in the first line, > is the proof real or claimed. classification, not prose. • what it found: 1,220 posts broke into the top 1%. baseline 1.22%. > superlative claim - 2.34% viral. 1.92x baseline > contrarian take - 1.59%. 1.31x > launch / tool drop - 1.46%. 1.19x and numbered lists, the thing everyone writes: 0.55%. below baseline. the most used hook is the least viral one. full stop. • what you are watching: left is the post under analysis, right is Jev answering 14 typed questions about it, each with a confidence score. the run stops at 20.4s because that is when Jev finished all 100k. pulled the corpus through a few X APIs, one parallel pass into Jev. should I drop it to public? Read my latest article on Jev Engineering below and turn your ideas into reality.

Movez

82,397 просмотров • 1 месяц назад

stop picking one model, a spider merged OPUS 5.5 + SONNET 5.5 + JEV + OPENAI DOTS into one business system one command splits a business into 6 jobs: plan, code, decide, run 24/7, review and ship every job has 4 candidates. the spider scans them one by one, marks the highest score as best, locks it in and swings to the next job the plan goes to Opus 5.5 at 0.94. Jev scores 0.52 on the same job on decisions it flips. Jev wins at 0.97 and Opus 5.5 gets 0.74, so the model that lost the plan is the one making the calls the code goes to Claude Code, and the job called "who never sleeps" goes to OpenAI Dots Opus 5.5 comes back a second time to check the work, and Sonnet 5.5 gets it out at the end 6 of 6 modules are locked and the system goes online run the same test on your own stack before you pay one model to do all six jobs ↓
0:41

stop picking one model, a spider merged OPUS 5.5 + SONNET 5.5 + JEV + OPENAI DOTS into one business system one command splits a business into 6 jobs: plan, code, decide, run 24/7, review and ship every job has 4 candidates. the spider scans them one by one, marks the highest score as best, locks it in and swings to the next job the plan goes to Opus 5.5 at 0.94. Jev scores 0.52 on the same job on decisions it flips. Jev wins at 0.97 and Opus 5.5 gets 0.74, so the model that lost the plan is the one making the calls the code goes to Claude Code, and the job called "who never sleeps" goes to OpenAI Dots Opus 5.5 comes back a second time to check the work, and Sonnet 5.5 gets it out at the end 6 of 6 modules are locked and the system goes online run the same test on your own stack before you pay one model to do all six jobs ↓

Noisy

11,722 просмотров • 28 дней назад

this must be f*ckn illegal... i pointed a crawler at my own prompt instead of the web, 195 words, every word crawled as a decision instead of text each word gets a typed question: is it relevant, sourced, ambiguous, or needs approval. each answer returns a probability. then the routing is brutal and it has no opinion about it → p ≥ 0.95 goes straight to code, no model call at all → p < 0.95 goes to sonnet 5.5, that is the only part a model thinks about → anything that smells like approval goes to me one crawl, 168 forks out of 175 words read. 133 ended inside an if statement. 24 reached sonnet. 11 came to me. the good part: "ask" scored 0.544 on needs approval, no confidence either way, so it stopped and handed it off instead of guessing. average confidence across the run, 0.925. everyone crawls the web for context. it was sitting in the prompt, unread.
0:20

this must be f*ckn illegal... i pointed a crawler at my own prompt instead of the web, 195 words, every word crawled as a decision instead of text each word gets a typed question: is it relevant, sourced, ambiguous, or needs approval. each answer returns a probability. then the routing is brutal and it has no opinion about it → p ≥ 0.95 goes straight to code, no model call at all → p < 0.95 goes to sonnet 5.5, that is the only part a model thinks about → anything that smells like approval goes to me one crawl, 168 forks out of 175 words read. 133 ended inside an if statement. 24 reached sonnet. 11 came to me. the good part: "ask" scored 0.544 on needs approval, no confidence either way, so it stopped and handed it off instead of guessing. average confidence across the run, 0.925. everyone crawls the web for context. it was sitting in the prompt, unread.

Defileo🔮

28,506 просмотров • 29 дней назад