Mukunda Johnson @2024specs

AI Engineer. Sharing the latest in new tech advantages. No marketing posts. USA Joined January 2024

Tweets

326
Followers

186
Following

155
Likes

336

Mukunda Johnson @2024specs

a year ago

Be wary of OpenAI's tactics. Test their new model on unsolved problems, not known problems. It's known that AI vendors are training on many prewritten solutions from paid labor. OpenAI states it is "a new training dataset specifically tailored for it" openai.com/index/introduc…

0 0 0 39 0

Mukunda Johnson @2024specs

a year ago

venturebeat.com/ai/googles-gem…

0 0 1 33 0

Mukunda Johnson @2024specs

a year ago

Other studies have shown that models will be okay if they retain the older data set within their training. I'd say that "model collapse" would appear more like bias, not nonsense. theregister.com/2024/07/25/ai_…

0 0 2 35 0

Mukunda Johnson @2024specs

a year ago

Be wary of AI benchmarks. AI vendors are paying people to fine tune their models to target common benchmarks and make their product more appealing. I personally got an offer on Upwork to solve programming problems for AI. It's not real-world work.

0 0 2 38 0

LlamaIndex 🦙 @llama_index

a year ago

And if you want to run hosted versions of the Llama 3.1 family, our friends at Fireworks AI have got you covered:

Fireworks AI @FireworksAI_HQ

a year ago

And if you want to run hosted versions of the Llama 3.1 family, our friends at Fireworks AI have got you covered:

8 18 109 22K 29

Download Image

2 5 20 5K 7

Mukunda Johnson @2024specs

a year ago

Meta released Llama 3.1 and suggests it is competitive with GPT4, 4o, and Claude 3.5 Sonnet. ai.meta.com/blog/meta-llam…

1 0 0 35 0

Mukunda Johnson @2024specs

a year ago

zdnet.com/article/apple-…

0 0 0 7 0

Mukunda Johnson @2024specs

a year ago

Before reading an article, scroll to the bottom and look for a section titled "Conclusion". If you see that, don't read it. People are trashing the internet with output from AI for a quick buck, and that will hurt AI's future. Don't give attention to these careless people.

0 0 1 24 0

Mukunda Johnson @2024specs

a year ago

This is a curious figure for this demograph reuters.com/technology/art…

0 0 0 10 0

Mukunda Johnson @2024specs

a year ago

60% cheaper than GPT-3.5 Turbo reuters.com/technology/art…

0 0 0 16 0

Mukunda Johnson @2024specs

a year ago

videocardz.com/newz/poll-show…

0 0 0 12 0

Mukunda Johnson @2024specs

a year ago

LLMs are a "generalization of everything". A major caveat to this is that they can't be clever, and there are many parallels between different topics that it will incidentally align. theguardian.com/technology/art…

0 1 0 45 0

Mukunda Johnson @2024specs

a year ago

A software engineer's discipline is all about achieving correctness. It's hard to picture a future where this mentally intense discipline is done automatically by technology that, by principle, has correctness problems. qit.software/will-software-…

0 0 0 24 0

Mukunda Johnson @2024specs

a year ago

Today's functionality is hinging on unethical data harvesting. wired.com/story/youtube-…

0 0 1 11 0

Mukunda Johnson @2024specs

a year ago

An interesting approach, using a special medium to transfer large amounts of data into an LLM more efficiently thestack.technology/microsoft-llm-…

0 0 0 13 0

Mukunda Johnson @2024specs

a year ago

Analysts were split over whether today’s AI investments would pay off in the next decade. cointelegraph.com/news/ai-artifi…

0 0 0 10 0

Mukunda Johnson @2024specs

a year ago

Unfortunately there is no regulation of content pollution. Everyone wants "in" and it's going to contribute to potential model collapse. pymnts.com/artificial-int…

0 0 0 7 0

Mukunda Johnson @2024specs

a year ago

The profitable use is obvious: information discovery. Use AI to find things. Let your customers find things on their own, just the same as your employees. There are other uses, but this easy value should be in every business by now. news.slashdot.org/story/24/07/11…