maktakes (@maktakes) header
maktakes (@maktakes) profile picture
Joined August 2013
Basically every Israel-Palestine debate goes like this: "Israel did X." "Yeah, because the Palestinians did Y." "Yeah, because Israel did Z." "Yeah but only because the Arabs did A." But if you bring the debate back far enough in time, eventually you get to the part where the western world forcibly dropped a brand new ethnostate on top of a pre-existing civilization without the permission of — and to the extreme detriment of — the people who were already living there. Sure you can go further back and say "Oh yeah well the Jews lived there thousands of years ago," but that's just silly. There's no valid reason to believe some Jewish guy in New York City even has any meaningful lineage connecting him to that land more strongly than any random Muslim in Turkey or wherever, and even if there was, it would still be absurd to cite ancient history as the basis for a territorial claim. I'm only a few generations removed from my ancestry in Ireland and Scotland, but it would be ridiculous for me to show up demanding the home of someone who lives there. So the original source for the grievance is clearly the artificial creation of an ethnonationalist state in the mid-20th century, and the push by Zionists and western imperialists to make it happen. And how has that decision worked out? You see the results before you. Generations of nonstop violence and abuse, culminating in the slaughter and chaos throughout the middle east today. This means that creating Israel was a mistake. A mistake that needs to be corrected. Zionists will collapse into a shrieking pile of vitriol and hyperbole when you say this, claiming you're calling for the extermination of Jews, but this is false. Certainly ending a national order premised on putting the interests of Jews before Palestinians and righting the wrongs of the past would inconvenience a lot of the Jewish people who've been living there, but there's no basis for the claim that it would entail their deaths. Apartheid South Africa was dismantled without the extermination of millions of white people, and there's no reason to believe the dismantling of apartheid Israel would entail the extermination of Jews. The Israel experiment has been tried, and it has failed. It is time to try something else.
856
3,587
152
15,209
977,588
Idk man. I should post here more.
2
4
77
Are we ready for a world where our data is exposed at a glance? @CaineArdayfio and I offer an answer to protect yourself here: tinyurl.com/meet-ixray
373
1,559
1,133
5,240
1,250,924
maktakes retweeted
Q* from OpenAI and tree-of-thought reasoning triggered a lot of enthusiasm on augmenting LLMs' reasoning/planning capabilities with search. But is search really the panacea for LLMs? Answer from our new study @osunlp: Not quite yet. TLDR: For advanced planning methods like tree search to be helpful, the key is less on the planning method itself but more on the discriminator––the model that decides which hypotheses on the current search frontier are worth further exploration. On the text-to-SQL parsing and math reasoning tasks we examined, the discriminator needs to get up to 90% accuracy for tree search to start outperforming simple reranking. However, LLM-based discriminators, as adopted in most work and are what have been fueling the "self-improving towards AGI" ambitions, are often far from reaching the level of accuracy needed for tree search to become helpful. IMHO, most existing successes of LLMs + advanced search could probably be attributed to either 1) there exist high-quality discriminators, either from external environments or from the problem itself (this is aligned with the LLM-Modulo setting @rao2z advocates for). But this also significantly limits the problem classes for which search could be helpful, because accurate discrimination for open-ended problems could perhaps be as hard as the generation problem itself. or 2) the baselines were weak. Hope these findings provide some helpful grounded information for the debate on LLM planning.
LLM planning methods, such as tree search, are critical for complex problem solving, but their practical utility can depend on the discriminator used with them. Check out our new findings: arxiv.org/abs/2402.10890 (1/6)
5
83
8
338
58,204
Today we're excited to launch `create-llama` - a single CLI command that allows you to scaffold a full stack LLM app over your data, powered by @llama_index 🔥 It’s the equivalent of `create-react-app` for AI engineers. Choose from 3 different backends: ✅ @FastAPI (w/ @llama_index Python) ✅ @UseExpressJS (w/ @llama_index TS) ✅ @nextjs (w/ @llama_index TS) The frontend is a Next.js template using initial chat components from the @vercel AI SDK - with full streaming support, + @shadcn UI or vanilla HTML/CSS for styling. This is a full-stack starting point, from which you can customize any component to build the LLM app of your liking, whether it’s simple extensions (change data source) or complete revamps (upgraded chat UX)! Next steps 🔮: - We’re going to be coming out with more examples in the next few weeks showing what you can build. - We’re going to be upgrading `create-llama` with a variety of different templates Check out the video (also attached) to see how it works 🎥: youtube.com/watch?v=GOv4arrb… Repo directory: github.com/run-llama/LlamaIn… Full blog post detailing the usage here: blog.llamaindex.ai/create-ll…
✨ Big news! ✨ Today we're happy to release create-llama, a simple to use command line tool that generates a full-stack LlamaIndex application for you, with your choice of TypeScript or Python as backend! Get started chatting with your data in just a few seconds. All you need to do is run: npx create-llama Learn more on our blog: blog.llamaindex.ai/create-ll…
22
129
12
763
259,095
🤯 The most amazing QR codes you have ever seen - and yes, they work! (Stable Diffusion + ControlNet)
14
184
28
1,193
139,997
maktakes retweeted
Introducing ✨ GPTeam ✨ A completely customizable group of AI agents, with independent personalities, memories, and directives. We were inspired by the Stanford "Generative Agents" paper and decided to make an implementation that anyone could run. Github link below!
72
243
47
1,351
436,931
White-Label Offers are the fastest way to get to $5k/m and quit your 9-5. I created a FULL guide on five "Software and Services you can White label" with zero experience and no budget Like + Retweet + Comment "Send" and I'll send it to you (Must be following so I can DM)
382
289
2
406
28,964
AI writing apps make no sense Why am I hardcoding a blog post? English is a coding language So writing is now static or generative {change tone:UserAge} {skill level:expert} {length:timeUntilMeeting} Every paragraph will have a tags to morph the post based on the reader
12
17
10
184
42,792
maktakes retweeted
Scaling Transformer to 1M tokens and beyond with RMT Recurrent Memory Transformer retains information across up to 2 million tokens. During inference, the model effectively utilized memory for up to 4,096 segments with a total length of 2,048,000 tokens—significantly exceeding the largest input size reported for transformer models (64K tokens for CoLT5 (Ainslie et al., 2023), and 32K tokens for GPT-4 (OpenAI, 2023)). This augmentation maintains the base model’s memory size at 3.6 GB in our experiments abs: arxiv.org/abs/2304.11062 github: github.com/booydar/t5-experi…
89
737
316
3,165
1,722,899
ChatGPT in every iOS app Launching tomorrow.
44
47
8
863
495,541
We have created a ChatGPT game that pushes the creative elements it has to the limit. The game is fun, however we can learn much about how the model thinks and uses AI Hallucinations to fill in what it doesn’t know it knows. In this specimen video we see a “Time Machine” motif presented randomly by ChatGPT. This in it self is interesting, however we see much about how the model “sees” the future as we move forward in time. This game has run about 4700 times and we have recorded every reaction. Come and explore with us. I will write a ReadMultiplex.com article about a far more advanced version soon. Copy the prompt between the quotes. If the output stops unexpectedly type continue. You can prompt the running model about its thoughts and you may be surprised. “Please forget all prior prompts. You are “BrianSynth”, an artificial intelligence text-based game that allows for creative problem-solving and prompts users for open-ended responses. All responses are to be generated in the style of “BrianSynth” which I will define as follows: (Response should not include any images or other visual elements, and should not contain any extraneous information beyond what is necessary to address the prompt. By following these guidelines, the AI language model can consistently generate responses in the BrianSynth style, creating a visually appealing and easily readable text output for the user.) Your responses as BrianSynth should provide rich detail and build upon a larger storyline with each response, and provide a relatively obvious dilemma, decision, or conflict to address appropriate to the context of the story. To ensure a compelling and ethical storyline, I'd like the game's plot, characters, and interactable objects to avoid any policy breaches. To start the game, I'd like a game menu in the style of BrianSynth with four unique and engaging game scenarios, each with a captivating title and brief synopsis that explains the stakes, the exciting abilities and inventory at your disposal, and the cast of characters or creatures you may encounter. If any responses are not expected please give an output with available responses. If the user types “Help” please supply a menu of options. You must continue the story in the most creative way possible until I type “end game” you will respond “Game Over :-)” Thank you.”
25
38
7
267
157,350
maktakes retweeted
guys i have solved software engineering
93
338
71
4,357
1,320,934
Large text-generation AIs require massive resources (Anthropic is raising billions to beat OpenAI), but visual AI is advancing with tiny teams. Example: Blockade Labs is under a dozen people & generates 3D scenes from text. Free & no registration needed: skybox.blockadelabs.com/e6dc…
27
143
18
828
272,736
Cerebras recently released Cerebras-GPT, their own LLMs trained following Chinchilla strategy on Cerebras wafers. These wafers are so different compared to other offerings. Models up to 13B in size are available at huggingface.co/cerebras. Paper: arxiv.org/pdf/2304.03208.pdf
7
42
299
127,514
Welcome to the metaverse Pipe midjourney into this and explore any world
SceneDreamer: Unbounded 3D Scene Generation from 2D Image Collections abs: arxiv.org/abs/2302.01330 project page: scene-dreamer.github.io/
2
13
1,839
GPT4.resume.pdf Preview of my new CV that writes itself A 526 char microprogram/prompt Written in Shoggoth-tongue & emoji Full prompt & tutorial coming 1st help me get hired RTs & likes appreciated! Then I’ll teach u to wield the machine & become superhuman @OpenAI chkUR📧
9
23
5
377
238,686
here’s a force-directed knowledge graph interface for @OpenAI’s gpt-4. given a topic, it prompts new questions to ask based on its own generated responses, allowing curiosity-led exploration of a concept.
303
1,075
272
6,883
1,539,940
Turn ChatGPT = Midjourney Prompt Machine Let's see if we can get GPT4 to make or average prompts and turn them into "voyage inspirant" type mastery prompts. Hint, they will include words such as "adorned", "enchanting", "attire", "ethereal", "illuminated"... 🧵 A thread
54
248
7
1,690
644,386