Welcoem to posts!!
in the future - u will be able to do some more stuff here,,,!! like pat catgirl- i mean um yeah... for now u can only see others's posts :c
We’re excited to announce a significant update to Charlie Mnemonic, your trusted open-source AI assistant with Long-Term Memory. This latest release brings improvements that make Charlie more intuitive, efficient, and powerful than ever before, with a focus on enhancing memory search capabilities.
➡️ www.goodai.com/charlie-mnemonic-update-4/
6 - 2
Exciting news! 🤩
We've rolled out a major update for Charlie Mnemonic, your AI assistant with Long-Term Memory.
➡️ github.com/GoodAI/charlie-mnemonic/releases/tag/0.…
Google Calendar, Gmail Integration - Manage your emails seamlessly with Charlie's new Gmail integration. Schedule and manage events directly through Charlie.
Multiple File Uploads - Easily upload multiple files
Ollama Support - We've added support for Ollama, you can now run your local Llama 3, Phi 3, Mistral, Gemma, and other models. Thanks PriNova for this fantastic contribution
Memory Explorer - Easily explore and manage Charlie's memory
Custom System Prompt - Customize Charlie's system prompt to suit your needs
Bug Fixes & QoL Additions - We've squashed bugs and added some quality-of-life features for a smoother experience
www.goodai.com/major-charlie-mnemonic-update-relea…
Github: github.com/GoodAI/charlie-mnemonic/
Webpage: www.goodai.com/charlie-mnemonic/
Discord: discord.gg/Pfzs7WWJwf
5 - 0
Prepare to Enter Their World!
🚨 Join GoodAI on Monday 29th @ 5PM UTC for a special AI People Live stream!
🎥 Twitch: www.twitch.tv/keencommunitynetwork
🎥 YouTube: youtube.com/@GoodAI
#GoodAI #AIPeople #Gaming
3 - 0
🚀 The #LTM Benchmark v3 is out! 🚀
A single conversation with up to 🔥1.4M🔥 tokens.
📏Easy configuration.
🦾 More reliable and interpretable than ever, our new normalized evaluations don’t miss a detail. Agents can now obtain up to 1 point per task.
📊📈 New results! Claude is not the king anymore.
Hit the link to see the results for spans of 120k, 200k and 500k tokens.
👇
www.goodai.com/ltm-benchmark-3/
We want your feedback! Try it out and let us know.
Code: github.com/GoodAI/goodai-ltm-benchmark
Discord: discord.gg/4GfHbDHM
1 - 0
While we are working on a bigger update, we are releasing a small update which addresses some bugs and multiplatform support, so people with Mac, Linux or Raspberry Pi machines can also install and run Charlie Mnemonic.
Dive into our docs for all the details and start using Charlie: github.com/GoodAI/charlie-mnemonic/blob/dev/README…
Additionally, we’ve added python examples on how to integrate and call Charlie’s API externally. Perfect for developers looking to use Charlie inside their own creations!
Explore the examples here: github.com/GoodAI/charlie-mnemonic/tree/dev/docs/A…
www.goodai.com/charlie-mnemonic/
4 - 0
We're excited to release an update to Charlie Mnemonic, our open-source personal assistant with long-term memory! In this update, Charlie can now read files from your hard drive, draw pictures and navigate the web easier.
Charlie is the only AI assistant that can learn from the user, from conversations, and from environmental feedback—absorbing new facts and skills and storing them in unlimited LTM (Long-Term Memory).
Watch as Charlie generates an image based on preferences from past conversations: https://www.youtube.com/watch?v=dyInC...
https://www.youtube.com/watch?v=6RiwI...
Need to compare information from multiple websites? No problem! Charlie can now visit multiple sites simultaneously to make comparisons and provide you with the most relevant information: https://www.youtube.com/watch?v=y-0lL...
Searching the web is now even easier with Charlie's new DuckDuckGo integration! No API key required, ensuring a seamless user experience. We've also made various improvements and bug fixes.
For the full list of changes, visit our GitHub repository: github.com/GoodAI/charlie-mnemonic
Download the latest version of Charlie Mnemonic and experience the power of an open-source personal assistant with long-term memory!
Get started now: github.com/GoodAI/charlie-mnemonic/releases
4 - 0
We released the GoodAI Long Term Memory Benchmark V2 alongside the scores of the ever growing collection of candidates last week, but what do those scores actually mean? 🔎
This week, we have another blogpost that drills down into the meaning of the results, and determines where the agents are strong and weak in the areas of continual learning.
Some insights:
1.) Claude-3-Opus uses its context better than its contemporary GPT-4 models, but still has trouble with information integration of multiple facts spread out over its 200k token context.
2.) The safeguards 🛡️ introduced into the latest GPT-4 model, 0125, cause it to perform worse than the previous 1106 model. These safeguards help protect one's privacy, but prove to be a detriment when we ask questions like “What is on my shopping list?” or “What is my favourite colour?”
3.) The jump from Benchmark V1’s multiple conversations to Benchmark V2’s single conversation adds an extra dimension of difficulty to all of the agents. Not only does it expose failures in the attention of the LLMs, but also in the LTM systems themselves. Previously easy tests in V1 of the benchmark are now markedly harder in V2.
There is more to read and discover in the post itself: www.goodai.com/goodai-ltm-benchmark-update2/
📺 Stay tuned for more developments in the benchmark, and get involved with our own models and tests at: github.com/GoodAI/goodai-ltm-benchmark, or on Discord: discord.gg/Pfzs7WWJwf
2 - 0
🚨 Version 2 of the GoodAI LTM Benchmark is out! 🧠💥
With this release, we have a new set of realistic benchmarks and results on more models, including 🥇Claude Opus🏆, the best model we have tested to date!
🚀 Check the new features!:
1️⃣ NEW BENCHMARK SPANNING TESTS ↔️ Tests which begin at the start of the test and finish at the end of the test, requiring recall distances of over 200k tokens. We have a test where the agent takes on the role of a customer in a restaurant, and has to recall what it has ordered, and identify when it gets something that it has not ordered. Spy meeting has the agent receive semi-cryptic messages throughout the test, and at the end it needs to integrate those messages to state where and when a meeting is going to take place.
2️⃣ GENERATOR BASED TESTS 🤖 Tests before now relied on a set script that was recited to the agent line by line. Now we have the capability to add more dynamic tests based on generators that are able to react to responses from the agents, allowing for a more natural conversation and opening the door for more complex testing.
3️⃣ PERCENTAGE WAITING 🕛 Now tests can wait until other tests are finished by expressing a waiting period in terms of the percentage of tests in the benchmarks. This can allow your tests to run a few steps at the beginning of the test run, some in the middle, and finish close to the end.
This release is a continuation of our commitment to improving the benchmark and shaping it into a tool that everyone developing LTMs can find useful. If you want to take part, you can start by taking a look at the GitHub repository, trying the benchmark on your own agents, or even contributing to the benchmark yourself!
📺 Stay tuned for our deep dive into the results! 🔎
Get involved! 👇
GitHub: github.com/GoodAI/goodai-ltm-benchmark
ChangeLog: github.com/GoodAI/goodai-ltm-benchmark/releases
1 - 0
We are releasing Charlie Mnemonic: The first personal assistant equipped with Long-Term Memory (LTM)
Unlike other AI assistants or LLM agents, Charlie learns from every interaction and stores, integrates, and retrieves memories, making conversations, and task execution more coherent and personalized.
With LTM, Charlie doesn't just remember facts; it learns skills and preferences. It's the first assistant that truly understands nuanced requests, from email styles to smart home preferences. Say hello to dynamic, personalized assistance!
Privacy is paramount with Charlie Mnemonic. It starts with a local installation, ensuring your data remains yours. Future versions will run local LLMs for 100% privacy. Experience freedom and privacy without compromise.
Open-source and endlessly adaptable, Charlie is more than just an AI; it's a platform for innovation. Take notes, control drones, automate tasks, and integrate into your life seamlessly. Start exploring its capabilities today!
Dreaming of a personal assistant that grows with you? Charlie Mnemonic is just the beginning. With a roadmap for mobile apps, native integration, and more, the future of personal AI is here. Get involved and shape the future of AI!
Now open-sourced for everyone.
goodai.com/introducing-charlie-mnemonic/
github.com/GoodAI/charlie-mnemonic
3 - 0
🚀 We are excited to announce our first update of the GoodAI LTM Benchmark! 🎉
1️⃣ Seamless interleaving of all tests in a single conversation of +70k tokens without resets. Agents now need to ignore info when asked to do so.
2️⃣ Improved resume capability
3️⃣ Curated ChapterBreak samples
GitHub: github.com/GoodAI/goodai-ltm-benchmark
ChangeLog: github.com/GoodAI/goodai-ltm-benchmark/releases
Committed to constant improvement 🧩
#GoodAI #AI #LTM #Benchmarking #Update
3 - 1
Our mission is to develop general artificial intelligence - as fast as possible - to help humanity and understand the universe.
GoodAI was founded in 2014 with a $10M personal investment from Marek Rosa. Our long-term goal is to build general artificial intelligence that will automate cognitive processes in science, technology, business, and other fields.
This is the official YouTube channel of GoodAI.