ofcourse

ofcourse, 10 months ago

You can absolutely self host LLMs. HELM team has done an excellent job benchmarking the efficiency of different models for specific tasks so that would be a good place to start. You can balance model performance for your specific task with the model’s efficiency - in most situations, larger models are better performing but use more GPUs or are only available via APIs.

There are currently 3 different approaches to use AI for a custom task and application -

1. Train a base LLM from scratch - this is like creating your own GPT-by_autopilot model. This would be the maximum level of control, however the amount of compute, time, and data required for training does not make this an ideal approach for the end user. There are many open source base LLMs already published on HuggingFace that can be used instead.

2. Fine-tune a base LLM - starting with a base LLM, it can be fine tuned for a certain set of tasks. For example, you can fine tune a model to follow instructions or use as a chatbot. InstructGPT and GPT3.5+ are examples of fine tuned models. This approach allows you to create a model that can understand a specific domain or a set of instructions particularly well as compared to the base LLM. However, any time that training a large model is needed, it will be an expensive approach. If you are starting out, I’ll suggest exploring this as a v2 step for improving your model.

3. Prompt engineering or indexing using an existing LLM - starting with an existing model, create prompts to achieve your objective. This approach gives you the least control over the model itself, but is the most efficient. I would suggest this as the first approach to try. Langchain is the most widely used tool for prompt engineering and supports using self hosted base- or instruct-LLM. If your task is search and retrieval, an embeddings model is used. In this scenario, you generate embeddings for all your content and store the embeddings as vectors. For a user query, you then convert it to an embedding using the same model, and finally retrieve the most similar content based on vector similarity. Langchain provides this capability, but IMO, sentence-transformers may be a better starting point for a self hosted retrieval application. Without any intention to hijack this post, you can check out my project - synology-photos-nlp-search - as an example of a self hosted retrieval application.

To learn more, I have found the recent deeplearning.ai short courses to be quite good - they are short, comprehensive, and free.

ofcourse, 10 months ago (edited 10 months ago)

I agree with OP that instances being closed any time is an issue that would need to be resolved fairly soon. A solution in my opinion would be the option to transfer user accounts across instances. This would help with an instance closing and eventually make the fediverse more stable.

A new user currently has a choice for joining from a number of instances but there is no assurance to ongoing existence for them. Along with that, afaik there is no way to transfer user accounts and data across instances. If a user can transfer their accounts and data, there will be less hesitancy to join a new instance, and user accounts and data can be distributed across more instances. This can also work in such a way that if a subset of user data does not meet the criteria for another instance, then that subset of data is not migrated (most likely a community based data filter).

Another issue is with the presence of same community/magazine in multiple instances (let’s say tech@lemmy.this and tech@kbin.that) which is frustrating for users since they need to track multiple communities for similar content and the same content is being copied to multiple communities. This should also be resolved by implementing account migration. We are already seeing that communities on certain instances are becoming the prevalent ones. This creates an incentive for the admin of those instances to not shut down. And if they did decide to shut down the instance, then the users can just migrate to another instance and the prevalent community will also get to keep all its data, just in the new instance.

ofcourse, 10 months ago (edited 10 months ago)

Eating at a deficit makes trying to gain muscle a slow process. If you would like to gain muscle faster, your strategy of going surplus seems right. Keep at it, you are doing the right things and it will show eventually! Additionally, here’s some broad suggestions in case some resonate with you and others.

Diets and regimens work differently for different people. So I would preface everything by - what works for me or someone else may not work for you and you’ll need to do some trial and error to figure out what would work the best for you.

I read somewhere that you count calories to manage weight and exercise to get into/maintain shape. So that’s my TL;DR.

• Weight - Weight will always be determined by calories in, calories out. Your metabolism, macronutrients intake, sleep behavior, activity level, genetics, mental health will all impact how much calories your body consumes in a typical day. When starting off, try to count the calories you are consuming as accurately as possible. Monitor your weight over a few weeks and find your average caloric consumption based on that. Then use this to determine how much calories you should be consuming to get to the desired weight within the desired period.
• Carbs - Avoid too many carbs and definitely avoid high glycemic index carbs as much as possible. This is anything with simple carbs that the body breaks up easily. They are great to give you quick boosts of energy but they also fluctuate your glucose levels, and make you feel hungry sooner. This makes it hard to eat within a caloric limit and gives the unpleasant sugar highs and lows. That said, carbs are a macronutrient so don’t eliminate them completely. In fact they may help when trying to push your body harder in your workouts.
• Proteins - Protein is essential for your body to create and grow muscle tissue. It’s hard to get enough protein for muscle growth just from food sources without messing up total calories so supplements help. Get 25% or more of your daily calories from high quality protein when trying to gain muscle.
• Diets - Some of the recommended diets are intermittent fasting (IF), keto, paleo, vegetarian, vegan, etc. These diets have impacts beyond just your body weight but I would not get into them for this post. For your question, the most important aspect of all these diets is that they make it easier to manage your caloric intake. All these diets would also generally help you get the right macronutrients. I would suggest reading about them, finding what resonates with you and giving some a shot.
• Muscle - Compound exercise weight training with exercises that work multiple muscles, like squats, deadlifts, presses, and chin ups are good for beginners. Form is more important than the weight. Some people can get into very good shape with calisthenics only. Try to avoid machines in the beginning. Weight machines work fewer muscles at a time and don’t contribute to improving balance. As you learn more about your body and the exercises, you can add and modify them to what suits you the best.
• Sleep - Get good sleep and maintain your mental health. Sleep is essential for your body to work its magic of building muscle. And, if any of the above is making you miserable, it will be hard to stick with it for the long term. What even would be the point of looking great if you don’t feel good!

Personally, I’ve never been able to gain muscle and reduce weight at the same time. This means my total muscle mass is not increasing however my body is getting more in shape (which is what I’m trying to do). For some people they are able to gain muscle mass despite losing weight especially if they are just starting out with weight training. Don’t be discouraged if you fall in the first group. Once you understand your caloric intake and body, you can modify your diet and exercise to make it work for your goals.

ofcourse, 10 months ago

The arguments by the company, OceanGate, in the article are so strange. Paraphrasing -

• We are choosing to NOT get our vessel tested by another agency because most accidents occur due to human error instead of mechanical. While their claim may have some soundness for traditional and well-tested equipment, it doesn’t mean mechanical failures cannot happen especially for completely new designs and materials.

• We are testing using this new testing device that we ourselves created and are not going to use other prevalent non-destructive tests (NDT). It is perfectly acceptable to have to design new instruments for testing new materials since existing NDT instruments may be unsuitable for a new material. However, there still needs to be instrument calibration and testing, preferably by an independent body or at least peer reviewed and published, to verify that it will provide suitable evaluations. The article doesn’t mention if any of this was done.

Really unfortunate about all the people on board. It looks like some of the folks were rich so I hope the company is sued into oblivion and the leadership gets jail time for criminal negligence.

ofcourse, 10 months ago

I’m glad you got it going.

Apple attaches your iCloud account when you share a shortcut and I’m not sure about their policy on a shortcut like this, hence the screenshot.

If you find a better website for bypassing paywalls, please post here.

ofcourse, 10 months ago

It used to but it hasn’t worked for me the last few months. The paywalled pages show the same content as is visible without the reader mode, including the text saying that I need to be logged in, turn off adblocker, etc.

ofcourse, 10 months ago

Thanks for sharing your insights.

I’m curious why instances offering free search are defederated? I would have guessed everyone wants better search. Is it because of privacy concerns or instances don’t want to be indexed or have traffic directed elsewhere?

I was hoping that if I index only for the purpose of embeddings (which would prevent recreating the original content) and only share urls to the content that it should eliminate privacy and traffic concerns.

I’m still in the process of understanding how and if this would work. It’s only a personal project at this stage but you are right cpu/gpu and vector stores would be things I’d need to consider.

ofcourse, 10 months ago

I have found luck with Ikea smart bulbs. They don’t need to be connected to the internet to operate.

Tp-link devices are notoriously bad about connecting to the internet. There is no way to operate them without an internet connection. On top of that, each device connects independently so even when you have a vLAN, there’s extra steps required each time a new device is set up.

ofcourse, 10 months ago (edited 10 months ago)

Throughline has been my favorite since it launched a few years ago. The hosts take a deep dive into the historical events leading up to topical events of the present weaving a thread through them, hence the name.

Some of the examples are the history of policing in the US and how capitalism became the dominant economic system.

I cannot recommend this podcast enough!

ofcourse, 10 months ago (edited 10 months ago)

Mental Illness Happy Hour by Paul Gilmartin, if you like a podcast that talks honestly about the struggles of mental health.

Paul interviews a different person each week and discuss their journeys on dealing with their mental health. Paul is also been very open about his struggles. It helps that he is a comedian and has a subtle but dark humor that I enjoy.

I also really like the short surveys that he reads and people have filled out on his website because they make me feel connected that I’m not alone.

ofcourse, 10 months ago (edited 10 months ago)

I am surprised some of the big ones haven’t been mentioned yet -

• Radiolab - Not really sure how to describe this podcast. It’s superb journalism at its core. They do both short and multi-episode long form about a variety of topics from science to history to current events. For example, how dinosaurs died when the asteroid hit earth, the story of a Guantanamo convict with the same name as the host, and how poorly computer databases are designed for names that are outside the norm.

• Planet Money - An excellent economics podcast where complex topics are distilled in fairly short episodes. They recently released a completely AI generated episode which was incredibly scary with how good it was.

• More Perfect - Everything the US Supreme Court

• Serial - One multi-episode series at a time about complicated criminal cases.

• What Roman Mars can Learn about Con Law - Started off during the Trump Presidency when tough questions about the US constitution are being asked given his penchant for pushing the legal boundaries and norms.

ofcourse, 10 months ago (edited 10 months ago)

Some other factors that I have noticed -

• Since most of the democracies determine the result based on first past the post (FPTP) or closely related voting system, the candidates only need to get 50% of the voting population to agree with them. They focus on populist policies that resonate with at least 50.1% of the population even if those policies will be detrimental to the remaining 49.9%.
• The opposition is not seen as strong enough to lead the country. This was the case in recent Turkish elections and has been the case in the last 3 Indian elections. Erdogan and Modi keep winning because people who don’t want to vote for them are not convinced by the other candidates’ abilities to lead the country. So many of the opposing people don’t vote at all or have their votes fragmented across multiple candidates in FPTP systems. That was and also remains the concern with Biden in the US.
• Once these leaders are in power, they actively suppress the voice of the minorities, by controlling the media and law enforcement, or by making it harder for minorities to vote and express themselves. This reduces the total voting population in favor of these leaders which again benefits them get past the 50% votes. Ultimately, we observe the vicious cycle of more power consolidation over time and more authoritarianism.

ofcourse, 10 months ago

Magazines can be set to collect threads based on tags, so Bins sounds cooler for that scenario as well. Excellent idea @BlackCoffee!

ofcourse, 10 months ago (edited 10 months ago)

Adding to this, last I checked, zigbee was the most popular open standard for this.
I had originally started with tp-link smart bulbs and realized that they did not support zigbee. After I set up pi-hole on my network, I also noticed that they have a lot of internet traffic with no way to turn it off.
So I ended up discarding tp-link and start from scratch again. I have been using ikea tradfri smart bulbs. They support zigbee so they can be used with any zigbee supporting gateway or hub. The other amazing thing about them is the ease of using them without connecting to the internet. You can buy ikea buttons for them which can be connected with multiple devices so I have one for each room and it makes it really easy to turn a room on and off at once.

tl;dr - start with devices that support an open standard like zigbee. That will not tie you down to any one brand and give you the most options for future expansions.

edit - looks like Matter may be becoming the new standard and it’s published by the same organization that published zigbee. So that may be a better option if you are starting out with home automation.