OpenAI Drama - Socoder

-=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- (c) WidthPadding Industries 1987 0\|659\|0 -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=-
Socoder -> AI Threads -> OpenAI Drama
Mon, 20 Nov 2023, 01:52
Jayenkai	OpenAI Drama Over the weekend... OpenAI fires Sam Altman. A few other staff choose to leave with him. Microsoft hires a bunch of them. https://www.theverge.com/2023/11/20/23968829/microsoft-hires-sam-altman-greg-brockman-employees-openai -=-=- ''Load, Next List!''
Mon, 20 Nov 2023, 09:43
Kuron	OpenAI tried to hire them back, but then more folks left and made clear they were following Sam into the new venture. ChatGPT is well loved by me, but looks like it will be dying and not progressing any more as somebody new will mess it up.
Mon, 20 Nov 2023, 09:59
Kuron	Crazy shit, and was changing several times per day and overnight, but hell, may have to let Windows finally start updating to get my hands on any new goodies Sam is bringing to the table. Satya Nadella may have just made what history will declare to be the biggest power play in the history of Microsoft! This will likely rejuvenate MS. Given how pissed Satya Nadella was reported to be, this is not surprising. Wow!!!
Mon, 20 Nov 2023, 10:03
Kuron	former Twitch CEO and co-founder Emmett Shear has been named interim CEO. ROFLMAO! The old company is as good as dead, this hire is the final nail in the coffin. Even if it is interim, this SOB will kill anything he touches, not to mention making it cost for all levels of access. If you have stuff stored in ChatGPT get it out now and save it!
Mon, 20 Nov 2023, 10:21
Kuron	505 of 700 OpenAI employees sent a letter to the OpenAI board demanding that all current board members resign and Altman and Brockman be reinstated, or they will likely leave to join Altman and Brockman at Microsoft. Wow, there will be nothing left of the company. The difference between Chinese or Japanese run companies and Western run companies, is when stupid shit like this happens, the board members responsible would have the decency to hang themselves or jump out of windows for destroying the company. There is no coming back from this. OpenAI no longer has anything to share and nothing new will be created. MS will likely buy OpenAI for pennies on the dollar just to avoid future lawsuits.
Mon, 20 Nov 2023, 10:31
Kuron	the letter says that the OpenAI board "informed the leadership team that allowing the company to be destroyed 'would be consistent with the mission.'"
Wed, 22 Nov 2023, 04:01
Kuron	So... 747 out of 770 employees walking if Sam is not back... No way for the company to survive unless the board was removed and everything was undone... So glad Sam is back, as unbelievable and bizarre as this whole mess has been. Military level clusterfuck that did not have to happen, but the board cost the company billions and billions of dollars in shareholder value in 48 hours. Fired, hell, they should have their heads chopped off. This is the kind of $$ loss that causes people to end up missing or have accidents. Glad the saga is over, for now. Sheesh. One of the most promising AIs ever almost got destroyed, for no apparent reason...
Thu, 23 Nov 2023, 15:16
Pakz	https://www.theguardian.com/business/2023/nov/23/openai-was-working-on-advanced-model-so-powerful-it-alarmed-staff I have been reading about this breakthrough that might caused the problems there for a couple of days. They have this Q* breakthrough that alarmed the staff. It was said it would cause all sorts of jobs to be lost everywhere I think they implied. It is a sort of math solving feature that is stepping again yet closer to the race of a full agi system. More news is sure to follow. Next year might be the year they really start scaring the ... out of more than just a board of directors.
Thu, 23 Nov 2023, 15:22
Kuron	Based on the names of some of the old board and the new board, the people do not have the intelligence or the background to remotely understand what they are trying to discuss.
Thu, 23 Nov 2023, 15:57
Jayenkai	That's ok, they can just ask ChatGPT..
Thu, 23 Nov 2023, 16:14
Pakz	I looked into that Q* thing. It is just some sort of version of the A* algorithm. But different. I have let chatGPT make me a version that uses 2d maps. But it took me a while to have it create a working version. (I can get used to coding like that though) No idea how the code actually would be able to scare people? html/javascript Q* pathfinding example by chatGPT <!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <title>Q-Learning Pathfinding</title> <style> canvas { border: 1px solid #000; } </style> </head> <body> <canvas id="mapCanvas" width="400" height="300"></canvas> <script> const canvas = document.getElementById('mapCanvas'); const ctx = canvas.getContext('2d'); const numRows = 6; const numCols = 8; const cellSize = 50; const map = [ [0, 0, 0, 0, 1, 0, 0, 0], [0, 1, 1, 1, 1, 1, 1, 0], [0, 1, 0, 0, 0, 0, 1, 0], [0, 0, 0, 0, 0, 0, 1, 0], [0, 1, 1, 1, 0, 0, 1, 0], [0, 1, 0, 0, 0, 0, 0, 0], ]; function drawMap() { for (let row = 0; row < numRows; row++) { for (let col = 0; col < numCols; col++) { ctx.fillStyle = map[row][col] === 1 ? '#000' : '#fff'; ctx.fillRect(col * cellSize, row * cellSize, cellSize, cellSize); ctx.strokeRect(col * cellSize, row * cellSize, cellSize, cellSize); } } } function drawPath(path) { ctx.strokeStyle = '#00f'; ctx.lineWidth = 3; for (let i = 0; i < path.length - 1; i++) { const [row1, col1] = [Math.floor(path[i] / numCols), path[i] % numCols]; const [row2, col2] = [Math.floor(path[i + 1] / numCols), path[i + 1] % numCols]; ctx.beginPath(); ctx.moveTo(col1 * cellSize + cellSize / 2, row1 * cellSize + cellSize / 2); ctx.lineTo(col2 * cellSize + cellSize / 2, row2 * cellSize + cellSize / 2); ctx.stroke(); } } // Initialize Q-learning agent const numStates = numRows * numCols; const numActions = 4; // 4 possible movements: up, down, left, right const qValues = Array.from({ length: numStates }, () => Array(numActions).fill(0)); // Training loop (simplified for demonstration) const numEpisodes = 500; const alpha = 0.1; // learning rate const gamma = 0.9; // discount factor for (let episode = 0; episode < numEpisodes; episode++) { let currentState = 0; // Start from the top-left corner while (currentState !== numStates - 1) { const action = Math.floor(Math.random() * numActions); const row = Math.floor(currentState / numCols); const col = currentState % numCols; let nextState; switch (action) { case 0: // Move up nextState = (row > 0 && map[row - 1][col] !== 1) ? currentState - numCols : currentState; break; case 1: // Move down nextState = (row < numRows - 1 && map[row + 1][col] !== 1) ? currentState + numCols : currentState; break; case 2: // Move left nextState = (col > 0 && map[row][col - 1] !== 1) ? currentState - 1 : currentState; break; case 3: // Move right nextState = (col < numCols - 1 && map[row][col + 1] !== 1) ? currentState + 1 : currentState; break; default: nextState = currentState; } // Reward for reaching the goal const reward = (nextState === numStates - 1) ? 1 : 0; // Update Q-value (simplified) qValues[currentState][action] += alpha * (reward + gamma * Math.max(...qValues[nextState]) - qValues[currentState][action]); currentState = nextState; } } // Visualize the map and learned path drawMap(); const startState = 0; const goalState = numStates - 1; // Find optimal path using the learned Q-values let currentState = startState; const optimalPath = [currentState]; while (currentState !== goalState) { const action = qValues[currentState].indexOf(Math.max(...qValues[currentState])); switch (action) { case 0: // Move up currentState = (currentState > numCols - 1 && map[Math.floor((currentState - numCols) / numCols)][currentState % numCols] !== 1) ? currentState - numCols : currentState; break; case 1: // Move down currentState = (currentState < numStates - numCols && map[Math.floor((currentState + numCols) / numCols)][currentState % numCols] !== 1) ? currentState + numCols : currentState; break; case 2: // Move left currentState = (currentState % numCols > 0 && map[Math.floor(currentState / numCols)][(currentState - 1) % numCols] !== 1) ? currentState - 1 : currentState; break; case 3: // Move right currentState = (currentState % numCols < numCols - 1 && map[Math.floor(currentState / numCols)][(currentState + 1) % numCols] !== 1) ? currentState + 1 : currentState; break; default: currentState = currentState; } optimalPath.push(currentState); } drawPath(optimalPath); </script> </body> </html> --v
Thu, 23 Nov 2023, 21:39
Kuron	First reference to Q* although not by name: https://openai.com/research/improving-mathematical-reasoning-with-process-supervision Good summary: https://www.ghacks.net/2023/11/23/what-we-know-about-openai-q-star/
Fri, 24 Nov 2023, 00:58
Pakz	Here a video by a Trekkie explaining it and more. View on YouTube