Quanah Gibson-Mount

Anthropic released another paper on their ongoing work towards interpretability of large foundational models like their Claude 3 family, built on previous work by many. Their work continues to be enlightening and inspiring. The findings in this work are generally similar to previous findings on smaller models but scaled up to Claude 3 Sonnet: 💡 they find large numbers of "features" corresponding to combinations of neuron activations and these features map to concepts that we recognize/understand: for example, features for transit infrastructure, particular people, concepts (like bias and "objects lacking sentience"). This might sound familiar from classical machine learning, but the interesting thing is that there is no single thing/neuron encoding each of these concepts, it's a particular combination of neuron activations that embeds these concepts in "superposition" in the network. 💡they observe a sort of "distance" of these feature activations that arrange themselves "near" semantically similar features. When they use their sparse autoencoder approach to learn more and more features they observe that the level of abstraction in the detected features becomes more granular, in ways that are interpretable and meaningful. 💡when they artificially fire these activation patterns during inference, it affects the output in ways that relate to the "feature". For example, asking Claude "what's the most interesting science" normally it responds Physics, but if you clamp the feature activations for the feature "Brain Science" to 10x their max values then Claude responds: Neuroscience. Ways to interpret _and_ influence the model. Great readable work [1] along with an interactive explorer of some of the "features" themselves [2] 🎂 There's an easter egg too: if you scroll in the feature explorer tool and try and select one of the "Randomly selected features from 1M" you get a message "if you want to see the rest of the features, we're hiring!" with a link to their recruiting page :) [1] https://lnkd.in/gZbparwp [2] https://lnkd.in/gC9eus53

19

"Recently, my PC's hard drive failed, leaving me with a cryptic error message and a looming deadline. With only 12 hours of sleep over 72 hours, I embarked on a troubleshooting odyssey. I delved into the world of bootloaders, file systems, and operating systems, determined to revive my PC. Through this journey, I discovered the importance of: 1. *Fault tolerance*: My PC's failure taught me that even the most critical components can fail, highlighting the need for backups and redundancy. 2. *Adaptability*: I learned to navigate unfamiliar territories, from Linux terminal commands to boot loader configurations. 3. *Resourcefulness*: With limited resources, I leveraged online forums, documentation, and creativity to diagnose and repair my PC. 4. *Resilience*: Despite setbacks and frustration, I persevered, driven by the desire to learn and overcome. This experience taught me that life can be a powerful teacher, often in unexpected ways. Just as a failed hard drive can reveal the importance of fault tolerance, life's challenges can reveal our strengths and weaknesses. In the end, I emerged with a dual-boot PC, a newfound appreciation for Linux, and a deeper understanding of the importance of adaptability, resourcefulness, and resilience. Life may throw unexpected errors our way, but it's how we respond that matters. By embracing challenges and learning from failures, we can become more robust, adaptable, and resilient – just like a well-designed system."

10

Project update! My application tracking extension is now doing fuzzier matching and injecting new information into LinkedIn (and other sites) to keep me from doubling up on applications. Next up: local storage and bulk import/export operations...

12

2 Comments

🌶🌶🌶 Hot take… Old problem: if we put more devs on the project, we can deliver it sooner. Same thinking, new manifestation: we can hire more overseas devs for less total money to produce the same result as our US staff. Flaw in that thinking: product development is not an assembly line. Devs are not “interchangeable”. Software is as much an art as it is science. It’s a team sport. The people on your team matters. #tech #staffing #management #hiring #jobs #layoffs #software #startup #teamleadership

25

6 Comments

TL;dr - RIP Terraform. Time to say goodbye. I just ran into this obituary for Terraform and Hashicorp. The most probable future seems to be the maximization of sales motion and the minimization of engineering investment - just adding another product to the price list and unleashing the sales force at it. The worst possible outcome would be if IBM decided to make Terraform part of their still-born cloud offering. Or maybe we’ll have Terraform for mainframe? #terraform #hashicorp #ibm #upgrade

32

5 Comments

In these times of economic uncertainty, why would you focus on database clustering? Here are a few reasons: ➡ High-risk industries like banking, finance, government, healthcare, and manufacturing face downtime costs of up to $85,000 per minute ➡ For enterprises heavily reliant on data centers, such as telecommunications or e-commerce, a single event's downtime cost can exceed $1 million ➡ The average cost of downtime across all industries has risen to about $9,000 per minute, with variations based on business size and industry ➡ Beyond immediate financial losses, downtime costs include lost productivity, recovery costs, and intangible costs like damage to reputation, loss of business, opportunity cost, poor user experience Clustering your database for high availability shields you from such hefty consequences. Henrietta Dombrovskaya has been kind enough to invite me to speak on this topic at the DRW Training Center as part of the Chicago PostgreSQL User Group meetup on Wednesday, 15th May. Attendance is free - I'd love to meet and discuss!

25

1 Comment

I've been testing GPT-4o since its recent release. While I haven't really had a use for the multi-modal (images, audio, and text) capabilities yet, the improvement in speed of response is noticeable. Not just noticeable from a spec sheet perspective either—it's noticeable in how the product works in real use cases. I've been thinking about why that is, and here's my theory: there's a breakpoint somewhere based on latency and response speed. Above this breakpoint, interactions feel sluggish and disruptive. Below it, they become seamless and effortless, akin to a natural conversation. Previously, I posted about how Llama and Gemini models seemed noticeably more usable than GPT-4, with my preference being Llama whenever possible. This new experience with GPT-4o further emphasizes the importance of optimizing both latency and throughput in AI systems. This chart from ArtificialAnalysis.ai shows that GPT-4o has a total response time of 1.5 seconds for 100 tokens, compared to GPT-4's 5.7 seconds. This speed difference isn't just a matter of numbers—it's fundamental to achieving a smooth and efficient user experience. When latency drops below a certain threshold, the interaction becomes fluid and engaging, making the AI more practical and user-friendly. It also makes its mistakes more tolerable. This improvement transforms how we use these models in real-time applications, making AI interactions faster and more intuitive. It underscores that achieving low latency and high throughput is no longer a nice-to-have but a critical requirement for practical AI deployment.

9

3 Comments

Here's how to export a ARRL Field Day Cabrillo log from WSJT-X: https://lnkd.in/gE5HEyVE

1

Familiarity Levels at Work, and Meeting Scheduling Novice: You are surprised by their calendar blocks Journeyman: You already schedule around their blocks Expert: You know why those blocks exist and when they are negotiable Elite: Already synced over chat

7

2 Comments

What was the last customer conversation you were part of? At Mish Guru we gave all of our engineers a challenge - go and watch 10 customer calls and write down something you noticed. The results weren’t what we expected. Engineers by and large are shielded from customers by layers of other people. Their main interaction is through second hand data in tickets and they often aren’t privy to the conversations that happen before features are decided on. The biggest shock to engineers when they watched sales or discovery calls was this: People actually like our product… That was when we realized the obvious. Engineers deal with failure each and every day. Bug reports, missing features, slow performance. They don’t think about the wins, they aren’t important to the job. They only see the flaws. If you only look at flaws day after day, you start to form a negative opinion of the product that you’re building. We started this experiment looking for product insights, but by breaking engineers out of the typical roles that we’d previously placed them in, we learned so much more about our own company. We actually got a heap of product insights as well, and I’ll write about those some other time. #productmanagement #engineeringmanagement #shapeup #agile

41

🚨 A Cautionary Tale for AWS Users: The Empty S3 Bucket That Cost $1,300 🚨 I recently came across an interesting story that serves as a critical lesson for those leveraging AWS services. Imagine setting up an S3 bucket for testing, expecting no charges because it’s empty. But then, you unexpectedly receive a bill of $1,300! 💸 Here’s what happened: Someone had unintentionally configured an open-source tool incorrectly, causing it to send backup requests to the wrong S3 bucket. This empty bucket, which was meant only for testing, ended up receiving about 100 million unauthorized S3 PUT requests in a single day. The key takeaway? AWS charges for unauthorized requests, even if they result in 4xx errors. Each failed attempt by the open-source tool to access the bucket was racking up charges on the owner’s bill. When an unauthorized request is made to your S3 bucket, such as: aws s3 cp ./file.txt s3://your-bucket-name/random_key you'll be charged for it, despite the AccessDenied error. Interestingly, no AWS account is needed to initiate such requests. Anyone who knows the name of any of your S3 buckets can ramp up your AWS bill as they like. #AWS #CloudComputing #S3Bucket #CostManagement #DevOps

10

Do you think you need to hire developers as w2 employees to build a highly-skilled, devoted and culturally aligned team? Let Blueprint show you otherwise. Here's how we're changing the game: 1️⃣ Unmatched Quality: Quality is our creed. Our vetting process includes interviewing with former CTOs and senior software engineers, ensuring only top-tier talent makes it to your team. 2️⃣ Long-Term Commitment: Forget the turnover and churn worries! Our developers stay longer than industry norms for W2 developers (average tenure of 2.2 years), providing stability and continued expertise to your projects. 3️⃣ Cultural Integration: Culture isn't just an add-on for us. We dive deep to understand your core values and ensure every developer we match with your team enhances and embraces your company culture. 4️⃣ Equity is Optional: Save your cap table for your growing startup! We ensure that our developers are passionate believers in your product right from the start - without incentivizing them with equity. We rigorously test for craftsmanship, ensuring they value not just a quality end result but also the journey of creation. 🌟 Shoot us a DM to connect and see how we can improve your staffing strategy. #StaffAugmentation #InnovationInStaffing #TechTalent #CultureFit #LongTermGrowth

20

1 Comment

There's so much that we can do together; I know at least a half-dozen open source companies all working in different parts of this space.  It's incumbent on all of us to help explain how the parts fit together into a whole, and to explain what that whole can do.

7

What makes a senior software engineer - senior? Lots of people can claim 5+/10+/30+ years of experience, or expertise in C#/JavaScript/whatever, but I've found that senior is much more an attitude than a number. After all, some people may have 20 years of experience, but have the same year of experience every year over and over again, 20 times. Senior engineers in my view typically: know the business cases for their features (at least) pretty well, do more with less, know when to ask for help, and have no problem saying "I don't know". What are some attitudes of a senior developer?

20

12 Comments

This. If you become a large organization and adopt a bunch of microservices, then K8s might be a great fit. That said, you really are just trading operational complexity at the end of the day. Where k8s really wins is in Google's use cases where you need to tightly pack and schedule workloads. K8s has a lot of really good functionality to move things around to get the most from your hardware. But if you just want to run a container, spin up a cheap VM, cp a docker-compose.yml, and call it a day.

12

4 Comments

Check out our next Expert Observability article by Kevin Scaldeferri on leveraging Custom Events. https://lnkd.in/gYTt-QJa #newrelic #observability #monitoring

22

While the corporate culture and return-to-office policies differ "markedly" between the three companies, the similar effects of the RTO mandates suggest that "the effects are driven by common underlying dynamics," wrote the authors of the study. "Our findings suggest that RTO mandates cost the company more than previously thought," said David Van Dijcke, a researcher at the University of Michigan who worked on the study. "These attrition rates aren't just something that can be managed away." Robert Ployhart, a professor of business administration and management at the University of South Carolina, said executives haven't provided much evidence that RTO mandates actually benefit their workforces. "The people sitting at the apex may not like the way they feel the organization is being run, but if they're not bringing data to that point of view, it's really hard to argue why people should be coming back to the workplace more frequently," Ployhart said. Senior employees, he said, are "the caretakers of a company's culture," and having to replace them can have negative effects on team morale and productivity. "By driving those employees away, they've actually enhanced and sped up the very thing they were trying to stop," Ployhart said. https://lnkd.in/gWntKQF2

1