Omich

Hello friends, my game is now available on itch.io. Please give it a try and share it with others if you like it. At this stage, I really need attention to the project 🙏 omich.itch.io/totally-normal… #gamedev #indiedev #itch #totallynormalgame

@bhutanisanyam1 and Orca paper tells us that we need to put in train dataset more information of every aspect of reflection, which really makes the model more meaningful. This approach to training, combined with tree of thoughts, can boost GPT-4 power by a lot.

@bhutanisanyam1 I wonder what if we train big models like GPT-4 on dataset about how people try to solve problems using a tree of thoughts? I mean "tree of thoughts" is the best current approach for LLMs

Progressive Learning to outperform ChatGPT! 🚀 Orca, new 13B model trained using Progressive Learning is extremely performant on a lot of benchmarks and gives a strong challenge to ChatGPT. Here are my notes: - The authors use the Flan Dataset to generate 5M responses from ChatGPT and GPT-3.5 - Progressive Learning: The model is first trained on GPT-3.5 responses and then fine-tuned on GPT-4 dataset - System prompts: Most papers to my knowledge have ignored system prompts - Orca paper utilities 16 different prompts to control the variations of responses, which is another secret sauce Following all of this, we end up with a model as performant as GPT-3.5 on many benchmarks. There’s one interesting insight: System Prompts, although are helpful with training, are not impactful during inference. Im curious if this is due to lack of clever prompting? It’s cool to witness more fine tuning techniques for models. I’ve summarised the paper above but I’d suggest checking out the benchmarks: arxiv.org/abs/2306.02707

@EssebSaccad Хорошо, в следующий раз буду просить даже если за это меня мучает совесть :^) Кстати с ludum dare я пролетаю, ибо у меня завал на учебе и работе. Тогда уже буду просить помощи на следующем твг, если не найду в чем поучаствовать пораньше.

@EssebSaccad Да как-то не хотелось беспокоить без серьезного повода. Конкурс, по правилам, позволял вставлять в игру не свою музыку. Думал взять более менее подходящую музыку с интернета, в итоге даже этого не успел, лол. Из ближайшего, если буду делать что-то на следующий ludum dare свяжусь.

@EssebSaccad TWG 13 на доске /gd/

So the #lowrezjam has come to an end. To my surprise there was a lot of feedback. Thank you all for rating and comments, this is very motivating. Congratulations to all who got a high place, it's really deserved.

#LOWREZJAM Pinky gameplay: youtube.com/watch?v=fmoMFi…

Release omich.itch.io/pinky #LOWREZJAM

There are still a lot of work. I hope I succeed. #LOWREZJAM

little progress #LOWREZJAM

@SBRLabs Understood, thanks. I use Game Maker and compile under Windows. HTML5 for some reason can't use some functions.

@kitty_fromhell I was only asking what you are using to build your game 🙂 For example, I'm using HTML5 and plain-old Javascript for mine.

@SBRLabs Sorry, I don't really understand English and I don't understand the "tech" that you want 😶

@kitty_fromhell Hey, awesome! I've been enjoying watching your progress as I've been building my entry. What tech are you using there? I love the particle system there for the pink slimy bits that come off the blob.

Look better, I think #LOWREZJAM

🍴 #LOWREZJAM

New interactions #LOWREZJAM

@straydogstrut7 @f_oatley Thank you 😄