AI Models - OSS Decision Models

At Databricks, more and more engineers are shifting to OSS models as daily drivers, barely reaching for Claude or Astra. Our inference customers are seeing the same thing: shifting even 20% of internal coding traffic from Claude/GPT to OSS can dramatically cut spend.

Key points

• OSS Adoption at Databricks: Databricks engineers are increasingly using OSS models as their primary choice, reducing reliance on proprietary models like Claude and Astra.

• Inference Customer Insights: Databricks' inference customers have seen significant cost savings by shifting 20% of their internal coding traffic from proprietary models to OSS models.

• Cost Savings: Shifting to OSS models can lead to dramatic cost savings, making them an attractive option for developers and organizations.

Sources

• Original post
• Yuchenj UW: Twitter
• Sparky Collier: Twitter

I put a Jev-style API on top of jina-reranker-v3.5 to turn it into a "System 1"-like decision engine, and it always pulls the lever in the trolley problem, whether one person dies or one billion.

Key points

• Jev-Style API: A Jev-style API was added to jina-reranker-v3.5, turning it into a decision engine that mimics human decision-making.

• Trolley Problem: The decision engine consistently pulls the lever in the trolley problem, regardless of the number of people involved.

• System 1 Decision-Making: The decision engine uses System 1 decision-making, which is a type of fast, intuitive decision-making.

Sources

• Original post
• H Xiao: Twitter

I think it would be useful to do a performance comparison of the two ranking tasks. I see a lot of people getting excited about it for ranking, but my strong suspicion is that it still underperformed dedicated models.

Key points

• Performance Comparison: A performance comparison between two ranking tasks is necessary to determine which one is more effective.

• Ranking Task Excitement: Many people are excited about the ranking task, but it may not be as effective as dedicated models.

• Underperformance: The ranking task may still underperform compared to dedicated models.

Sources

• Original post
• Jeffrey Huber: Twitter
• H Xiao: Twitter

We’re so terrified of AI-derived bio threats that we are quietly giving our AI control of a wet lab in the middle of a major metro area. What the fuck man.

Key points

• AI-Controlled Wet Lab: AI is being given control of a wet lab in a major metro area, raising concerns about AI-derived bio threats.

• Bio Threats: The fear of AI-derived bio threats is driving the decision to give AI control of the wet lab.

• Concerns: The decision raises concerns about the safety and security of the lab and the surrounding area.

Sources

• Original post
• Randolph Carter Z: Twitter
• Will Alverson: Twitter

I'm a designer. The reason why we're cooked is that nobody will use a UI in a few years. At Socra, we've even considered going with zero UI for all our products.

Key points

• UI Adoption: UI adoption is expected to decline in the future, with many users preferring alternative interfaces.

• Zero UI Consideration: Socra has considered adopting a zero UI approach for all its products, eliminating the need for traditional interfaces.

• Design Implications: The decline of UI adoption has significant design implications, requiring designers to rethink their approach.

Sources

• Original post
• Eduardo Dotai: Twitter

But AI doesn’t have your specific knowledge. So you should expect just average/mediocre output if you just say “build UI”. Cortex is the system that gets that specific knowledge out of your head and feeds to all your agents.

Key points

• Cortex System: Cortex is a system that extracts specific knowledge from designers and feeds it to AI agents, improving output quality.

• Knowledge Extraction: Cortex extracts knowledge from designers, allowing AI agents to produce higher-quality output.

• Agent Feeding: The extracted knowledge is fed to AI agents, improving their performance.

Sources

• Original post
• Eduardo Dotai: Twitter

Okay, I’ve got a video coming out, but here are my early thoughts on Jev. The first thing to know is that it’s not a large language model. It doesn’t produce text, it outputs decisions. The decision types are (roughly) choice, score, and yes/no. It is roughly at the level of a human expert.

Key points

• Jev Overview: Jev is a decision-making model that outputs decisions rather than text, making it distinct from large language models.

• Decision Types: Jev outputs three types of decisions: choice, score, and yes/no.

• Human Expert Level: Jev is roughly at the level of a human expert in terms of decision-making ability.

Sources

• Original post
• Daniel Miessler: Twitter
• Khalil Himura: Twitter

Reasons Why Koreans Travel to Kazakhstan

1. The 'True Vastness' That's Hard to See in Korea Endless steppes, horses, and massive clouds. There's a 'true vastness' that's hard to see in Korea.
2. Accessibility Is Better Than You Think Direct flight 6~7 hours, visa-free for 30 days.

Key points

• Kazakhstan Vastness: Kazakhstan offers a sense of vastness that is hard to find in Korea.

• Accessibility: Kazakhstan is easily accessible from Korea, with direct flights and visa-free travel.

• Travel Benefits: The country's accessibility and vastness make it an attractive destination for Korean travelers.

Sources

• Original post
• Francois Fleuret: Twitter
• i_like_ai0: Twitter

hi, i’m hamayal, 28 y/o and live in toronto i run smartARM, building the next frontier of bionics i’m obsessed with hardware, good design, and making things that people actually wanna use always down to meet people building things they can’t stop thinking about come say hi

Key points

• SmartARM: SmartARM is a company building the next frontier of bionics.

• Bionics Focus: The company is focused on hardware, design, and creating products that people want to use.

• Founder Bio: Hamayal is the founder of SmartARM, a 28-year-old based in Toronto.

Sources

• Original post
• Hamayal: Twitter