HNHacker News
TopNewBestAskShowJobs

airocker

278 karma · joined August 24, 2018

submissionscomments
airocker··on Run0, a systemd based alternative to sudo, announced
Or Kubernetes. Thats where a standard way of authentication/authorization should be there.
airocker··on Run0, a systemd based alternative to sudo, announced
NixOS at it again :)
airocker··on Run0, a systemd based alternative to sudo, announced
I have seldom come across unix multiuser environments getting used anymore for servers. Its generally just one user on one physical machine now a days. I understand run0's promise is still useful but i would really like to see the whole unix permission system simplified for just one user who has sudo access.
airocker··on Anthropic's Haiku Beats GPT-4 Turbo in Tool Use
But what if there is a bad model that fine tunes with only tools?
airocker··on Anthropic's Haiku Beats GPT-4 Turbo in Tool Use
Are tools fine tuned in both or just prompt based? For a fair comparison
airocker··on DBRX: A new open LLM
is this also the ticker name when they IPO?
airocker··on Ollama now supports AMD graphics cards
Thanks! Great to know. I did not know llama.cpp could do this. It should be pretty straight forward to support, not sure why they would not do it.
airocker··on Ollama now supports AMD graphics cards
Thanks! This is great to know.
airocker··on Ollama now supports AMD graphics cards
OS was possibly much more complicated to write at that time than CUDA is to write today. And competition is too strong. It might be more briefer than even Sun.
airocker··on Ollama now supports AMD graphics cards
yes, not sure you can do better than that. You cannot still have one instance of LLM in (GPU) memory answer two queries at one time.
airocker··on Ollama now supports AMD graphics cards
More than one is easy: put it behind a load balancer. Put one ollama in one container or one port.
airocker··on Ollama now supports AMD graphics cards
I heard "Nvidia for LLM today is similar to how Sun Microsystems was for the web"
airocker··on Why time seems to pass faster as we age
According to this book: https://valsec.barnesandnoble.com/w/slow-down-time-the-power... , it basically happens because the number of snapshots you encode per second goes down. Snapshots are visual cues that we encode. The best example is that if we are watching a fan start, after a certain speed, we cannot tell how many wings are there on the fan. We are not snapshotting fast enough. Younger people can probably tell the number of wings on a faster fan :). Various things slow down this snapshotting in our brain, and the best antidote is to get in the zone and do things with immense attention.
airocker··on Time is encoded in the weights of finetuned language models
I think a more general way to think about it would be to add any data and reduce weight. For eg, if we want to create geography vectors, we would add all geography data to fine tune and then take a difference. Now add this to any other model with same architecture, and you have a geography capable llm.
airocker··on More product, fewer product managers
Even if they have the time, my rule is that nobody who does not understand the complexity of the solution (not the problem or the what) should be allowed anything more than watch rights in jira and standups. I would love to hear one non coding product manager who does not run jira in their org of more than 100 people.

The problem is that it is hard to be a product manger and own those tough customer metrics. It is easy to do jira and get credited for engineering work.

airocker··on More product, fewer product managers
We are not talking about a good manager here. We are talking about the one who can program manage to the level of lowest jira issues.
airocker··on More product, fewer product managers
Why not also be the developer if you can do that much already?
airocker··on Kubernetes Needs an LTS
Two years may or may not be enough "long term" depending on the product and the business.
airocker··on Kubernetes Needs an LTS
minikube does exactly that! Runs new versions of kubernetes inside a single Docker container.
airocker··on Kubernetes Needs an LTS
Maybe GKE and EKS should make LTS versions.
airocker··on Kubernetes Needs an LTS
Our cluster has reasonable complexity and yamls work well so far. Maybe we will see what you are saying if we scale further. It allows us to maintain a much more complex cluster with fewer developers. And the software is well tested/reliable.

But not having LTS is really difficult. It is impossible to keep rewriting things. And all related components: terraform, helm, GKE etc also update too quickly for us to manage with a small team. We do some lower level infrastructure work so these problems bites us harder.

airocker··on Kubernetes Needs an LTS
This is sorely needed. It causes service disruption for weeks because of weird upgrade schemes in GKE (or any other vendor) which works on top of Kubernetes. There are too many options with non-intuitive defaults related to how the control plane and the node pools will be repaired or upgraded. Clusters get upgraded without our knowledge and breaks service arbitrarily. Plus, if you are doing some infrastructure level changes, you have to put in extreme effort to keep upgrading.

IMHO , infrastructure is too low level for frequent updates. Older versions need LTS.

airocker··on Indians Are Entering the U.S. Illegally in Record Numbers
(Increased influx === Modi is bad) is just plain untrue because the state of Punjab's police was never under Modi. In fact, separatist supported government has been in power in the most recent elections. Maybe US and Canada should investigate if this separatist movement solely exists so that people can enter the US and Canada. In 1984, there were massive riots and many sikhs got to immigrate to the west. It was followed by demands of a Sikh state. Now, we are not sure this movement even exists in India. India is not Europe where an independent land locked state can exist with limited resources and Pakistan as its western neighbor. This is just a Schrodinger Cat experiment where immigration policies of the west actually make the movement exist. Also, this reads like a full propaganda piece which even promotes a Shah Rukh khan movie. Maybe not worthy of an HN first page post.
airocker··on Ask HN: Is there a way to learn Neural Networks/LLMs through problem solving?
I can write some code and follow along with the authors. But is there a resource which would ask for questions and check/develop the learners intuition?

Like given a network what do you think will happen if we change sigmoid to Relu? And then actually change it and show the result?

Our Physics books in high school had pretty nice and challenging questions. I am not sure the best place to access those for ML/AI.

airocker··on Nvidia’s AI supremacy is only temporary
Inference will not work on cpu, I see 100 percent cpu frequently when doing cpu inference and we had to change to use GPUs for a large client.
airocker··on Modern genetic data suggests pre-humans were a group of only 1,280 individuals
Sounds like the number of employees at OpenAI in 3 years.
airocker··on My Caste
I am willing to bet the author can walk into any IIT as a professor.
airocker··on Ask HN: Why did Python win?
I had the fortune of working in a ruby heavy company after learning Python. I thought ruby had too many exception to rules compared to Python, python was more elegant. Every other win I think is a derivative of being a better language which in turn came from being a better community.
airocker··on Oregon decriminalized hard drugs – early results aren’t encouraging
extreme freedom leads a democracy to tyranny -- socrates
airocker··on Kata Containers: Virtual Machines that feel and perform like containers
Is it possible to add kata containers in eks or gke?
← PreviousPage 3 of 10Next →