About
Artist. Technologist. Autodidact. Punk Rock Enthusiast.
Articles by Benjamin
Activity
9K followers
-
Benjamin Rust reposted thisBenjamin Rust reposted thisllama.cpp just hit 100k stars on GitHub 🔥 Here's a bold prediction: within 18 months, 90% of all AI agents will be running locally on your own machine. Here's why local agents are inevitable: → Models keep getting more efficient. MoE architectures like Qwen3.5-35B-A3B give you frontier-level quality while only activating 3B params at inference. → Quantization keeps getting better. The work from Unsloth AI among others is outstanding. → Apple Silicon and MLX made local inference fast and accessible. 64, then 128GB of unified memory will go a long way. → Agents don't just need 200 tok/s. They also need reliable tool use, good reasoning, and long context — all things local models are getting great at. Your data never leaves your machine. No API calls, no rate limits, no vendor lock-in. 2026 is the year of Local Agents.
-
Benjamin Rust shared thisLarger slugs is something customers like me have been needing when deploying ML workloads. Using docker is an option, but it’s nice to be able to deploy a model without the extra step, especially if you have a school project that quickly needs a home. I’m not too worried about slower builds, develop, train, test, hit push heroku main. https://lnkd.in/eUHY_DctBigger Slugs and Greater Build Timeout FlexibilityBigger Slugs and Greater Build Timeout Flexibility
-
Benjamin Rust shared thisHad a long talk over dinner last night about supply chain attack and how to mitigate or deal with them at scale. Unless planned for, this is only going to get worse for teams that aren’t keeping an eye on all the moving parts of there workflows. I’m not certain about the correct approach, I have a good idea how to solve this on an individual basis, but at scale this could be nearly impossible to combat.Benjamin Rust shared thisThe OSS ecosystem is under attack, and it all started with a few innocuous options in Github. Let me explain. On Feb 27th, a new Github account called hackerbot-claw and opened a pull request on Trivy and immediately closed it. But it was too late, the Github Actions were already running. Github has an *incredibly dangerous* setting that allows forks to run Github Actions with the secrets and context of the upstream repository. This makes it trivial to export secrets. One repo, not that big of a deal right? Wrong, the real hack was just beginning. March 19th the attackers publish compromised releases of Trivy. Trivy is a popular open source security scanner. You know where that's used a lot? In Github Actions. Suddenly, everyone who was configured to use the latest Trivy version in their Github Actions had all their tokens compromised. A lot of those repos published NPM packages. Do you see where this is going? The attackers now started publishing compromised NPM packages. Those are pulled into CI workflows and onto developer machines, where they helpfully install a service on every machine to propagate the worm and pull instructions from distributed dead drops that are almost impossible to take down. CannisterWorm is 6 days old. 140 compromised NPM packages. Hundreds of thousands of downloads. Anyone who thinks this is over, isn't paying attention. Update: This post originally linked the compromise to an incorrect settings panel in Github org permissions dealing with private forks. In fact, the attack vector was a misconfigured pull_request_target within a Github action. Removed here https://lnkd.in/gaRDt7SS
-
Benjamin Rust posted thisSo in celebration of international Mathematics day - March 14, or 3/14 or 3.14, I did just that. I spent the last several hours doing Linear Algebra. Not so much for fun, but for class, but it seemed apropos to be mathing on math day. Go crazy today do some 10 x 10 matrix multiplication by hand for fun. 🤓
-
Benjamin Rust shared thisAll the posts claimed Claude solves decades old math problem. In fact it did not solve it ex nihilo it needed Professor Knuth to know what to say, how to say it and where to be creative. As good as Claude is, it still doesn’t create, it pattern matches and takes creative guidance from the human. It’s a tool, a platform, a coworker with a scratch pad, but it ain’t Al-Khwarizmi.Benjamin Rust shared thisNo, Claude DID NOT "solve the problem Knuth has been working on for 30 years" these kind of posts are just the usual hyped and perverse magnification of the real story. What Claude actually says ( full script in comments) #claude #debunks #hype Could Claude have gotten there without human guidance? Claude answer (excerpt): ### Almost certainly not, and the paper is quite candid about this. Several forms of human intervention were essential. ( ... ) it functioned more as a very powerful search engine guided by an expert human collaborator than as an autonomous mathematician. Knuth frames it as a "dramatic advance," but the paper's details reveal a human-AI collaboration where the human's steering was indispensable. #### So, guys, don't buy similar posts popping everyday in particular if they come up with yet another smashing evidence of AI intellectual superpowers, they are most likeky manipulation. No AI will replace humans in real thinking anytime soon.Claude Opus 4.6 Surprises Turing Award Winner, Solves Problem He’d Been Working on for Weeks | Analytics India MagazineClaude Opus 4.6 Surprises Turing Award Winner, Solves Problem He’d Been Working on for Weeks | Analytics India Magazine
-
Benjamin Rust posted thisOne of the things I love about the modern age is that there are no limitations on learning. I am taking a Mathematics of AI course next term. This requires understanding of Linear Algebra, Calculus, Probability and Stats. Probability and Stats are covered I just took classes on those, but LA and Calc are in my dsitant past, so I went to Udemy found two classes that covered Linear Algebra and Calculus 1 - 3 (differential, integration and vectors) and while traveling for work watched them back to back. Now I feel ready to go into next term with the appropriate understanding of the how, why and what of the math of AI. There is no excuse in this day to not continue to learn. The brain is like a muscle, you gotta work it to keep it functioning optimally.
-
Benjamin Rust posted thisI'm excited to share that I've been chosen as a Pro Bono Fellow in the Salesforce Accelerator – Agents for Impact. My primary goal as a Fellow is to to help nonprofits build trusted, human-centered AI solutions. One of my favorite things about working for Salesforce is the breadth of companies I meet. I was a part of the Salesforce.org team for a long time and non-profits are a big part of how I spend my time. In the last year I've worked with two at great depth to help them design AI solutions for their Environmental Impact organizations. I donate my time to help these teams learn to use technology to remove friction from their processes, sometimes with AI and sometimes not. As a Fellow I'll be able to help even more organizations with their projects.
-
Benjamin Rust reposted thisBenjamin Rust reposted thisOver the past month in corporate America, a few things have really stood out to me, especially stepping into an Account Manager role at Apex Systems. The fundamentals matter more than people think. Clear communication, managing priorities, following through, and asking the right questions are not just classroom concepts. They are the foundation of strong client relationships. Consistency, responsiveness, and attention to detail build trust over time. I’ve also been reminded that every company and every industry has its own language. You are not expected to know everything on day one. What matters is how quickly you listen, adapt, and sharpen your understanding. In account management, that means understanding what drives value for your clients and aligning your support to that. And maybe most importantly, being coachable accelerates growth. Seeking feedback, applying it, and retaining it builds credibility quickly. The people who grow the fastest are the ones who stay curious and intentional about improving. Still refining. Still improving. Focused on delivering value and building long-term partnerships. #Sales #ApexSystems #IT
-
Benjamin Rust shared thisAn article I wrote about a proof of concept I built. Using a network isolated deployed RoBERTa model. Kind of a cool project. Heroku - infrastructure for network isolated hosting of model Salesforce - ticket handling and triage Agentforce - next actions based on distilBERT analysis Hugging Face - Open Sourced distilBERT machine learning modelBuilding a Self-Improving ML Risk Assessment SystemBuilding a Self-Improving ML Risk Assessment System
Benjamin Rust commented on a post
2d
If we just keep kicking them they will decide in their own not to participate OR they’ll develop defenses and shit will get real, real quick.
Benjamin Rust commented on a post
1w
There is so much truth to this. Everyone claims to know that messy data is a problem, many claim to want to clean it up, but almost no one wants to pay for hassle to get it done. Cleaning data is not trivial. 90% can be automated, maybe more, but the last 10% or so is always going to be tough. It is this last 10% that is likely key to project success, campaign funding, etc. But mostly it'll just be the thing that considered an outlier on a report because most teams don't do the hard work up front.
-
Benjamin Rust reacted on thisBenjamin Rust reacted on thisSixty-three issues closed across thirteen projects in one day. One person steering all of it. The hardest part wasn't building. It was keeping up with what I'd already built. The machines consumed every task I queued and sat there waiting. The bottleneck didn't disappear — it shifted straight to me. To judgment. To deciding what matters next. Automation is extraordinary at doing more of what you already know needs doing. It's terrible at noticing what you haven't thought to look for. https://lnkd.in/emx8RhUD #AI #FutureOfWork #HumanJudgment #Automation #TheWorkOfBeing
-
Benjamin Rust reacted on thisBenjamin Rust reacted on thisllama.cpp just hit 100k stars on GitHub 🔥 Here's a bold prediction: within 18 months, 90% of all AI agents will be running locally on your own machine. Here's why local agents are inevitable: → Models keep getting more efficient. MoE architectures like Qwen3.5-35B-A3B give you frontier-level quality while only activating 3B params at inference. → Quantization keeps getting better. The work from Unsloth AI among others is outstanding. → Apple Silicon and MLX made local inference fast and accessible. 64, then 128GB of unified memory will go a long way. → Agents don't just need 200 tok/s. They also need reliable tool use, good reasoning, and long context — all things local models are getting great at. Your data never leaves your machine. No API calls, no rate limits, no vendor lock-in. 2026 is the year of Local Agents.
-
Benjamin Rust reacted on thisBenjamin Rust reacted on thisCan we collectively agree to stop bringing these things to conferences please? They get in the way and they have no relationship to your product, we already know you became an “agentic” company overnight, just like everyone else in the room did. You don’t need an expensive useless toy to tell us that. I know people will say “they’re great for grabbing people’s attention”, but seriously are we that easily distracted by shiny toys?
-
Benjamin Rust liked thisBenjamin Rust liked this𝟱 𝘀𝗶𝗴𝗻𝘀 𝘆𝗼𝘂𝗿 𝗱𝗲𝗮𝗹 𝗶𝘀𝗻’𝘁 𝗴𝗼𝗶𝗻𝗴 𝘁𝗼 𝗰𝗹𝗼𝘀𝗲 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗱𝗼 𝘄𝗵𝗶𝗹𝗲 𝘁𝗵𝗲𝗿𝗲’𝘀 𝘀𝘁𝗶𝗹𝗹 𝘁𝗶𝗺𝗲 After years in complex B2B sales, I’ve learned to recognize the signals before the actual “no” shows up. What helped me systematize this was The Challenger Sale, a book that became a core reference in my process. Dixon and Adamson break the illusion that relationships and constant follow-ups are what close deals. They don’t. What actually drives results is constructive tension, insights that challenge the client’s thinking, and taking control of the process. Based on that, here are five key signals: 1. You’ve stopped teaching something new If your last three interactions were just “alignment” calls and “status updates,” you’ve lost your position as the challenger. The client no longer needs you to think and if you’re not bringing insights, you become a commodity. What to do: Bring a real provocation. Share a market data point that challenges their assumptions. Ask a question they haven’t considered yet. 2. You’re only talking to people who already like you You may have an engaged manager, but you’ve never reached the CFO, CTO, or the board. The Challenger Sale calls this “multi-threading” and not doing it is a clear sign your access is being filtered. What to do: Explicitly ask to be introduced to other decision-makers. If your contact resists, that’s a critical signal about the deal. 3. The client has gone quiet on decision criteria Early on, they talked about ROI, implementation timelines, integration with system X. Now, when you bring it up, they deflect or say, “we’ll look at that later.” That usually means the criteria have changed and you don’t know how. What to do: Reopen the conversation around what matters now. Be ready to hear that the scenario has shifted. 4. You’re proposing without controlling the process You sent the proposal. You waited. You followed up nothing. The Challenger Sale emphasizes taking control of the sale, not aggressively, but by defining clear next steps with real client commitment. If every meeting ends without concrete action on their side, you’re just waiting on the sidelines. What to do: End your next meeting with a specific commitment from them, a date, an owner, a deliverable. 5. The pain you were solving is no longer urgent There was a clear pain point at the start. Now the client talks as if the problem is “manageable for now.” That usually means one of three things: they solved it another way, priorities shifted internally, or the budget disappeared. What to do: Ask directly: “What’s changed since our first conversation that made this less urgent?” The answer will tell you whether there’s still a deal or if it’s time to ask the hardest question of all. If you’re working in complex sales and feel stuck in a deal that isn’t moving (or a pipeline that looks full but isn’t converting) reach out. This is exactly the kind of situation I help solve.
-
Benjamin Rust liked thisBenjamin Rust liked thisI built an AI pipeline, knew every piece of it, and it still broke in three places when I moved it to a new project. Not logic bugs. Assumption bugs. Now imagine someone who never built it. Never triaged an issue by hand. Never debugged a failed deployment. They wouldn't even know where to look. That's the gap nobody's talking about. Agents can generate entire systems faster than most developers can read the output. But speed without judgment isn't productivity. It's risk. You cannot responsibly automate what you cannot do manually. https://lnkd.in/eGBFiKdh #AI #Leadership #SoftwareEngineering #AIAgents #Judgment
-
Benjamin Rust reacted on thisBenjamin Rust reacted on thisI'm sharing my open class on pre-training, post-training, and serving large language models: https://lnkd.in/egFY9dqg , Training and Deploying Large-Scale Models. This is the first year I have taught it, at MVA, and it is designed to bring together both machine learning and systems perspectives. I believe it's important to share this knowledge widely and make this technology more openly accessible. I also hope others will reuse these materials for their own classes or for their own learning!
-
Benjamin Rust reacted on thisBenjamin Rust reacted on thisSalesforce AI Research is hiring Senior MLE! If you’re excited about building agentic AI systems, working at the intersection of research and engineering, and turning frontier ideas into real product impact, this is a great role to explore. You’ll work with a strong cross-functional team across research, engineering, and product to help shape the next generation of AI platforms. Apply here: https://lnkd.in/gR4apghJ #Hiring #AI #MachineLearning #AgenticAI #Salesforce #AIResearch
Experience & Education
-
Summer Springboard
View Benjamin’s full experience
See their title, tenure and more.
or
Licenses & Certifications
Volunteer Experience
-
Coach and Mechanic
Allatoona Area Composite Mountain Bike Team
- Present 10 years 10 months
Health
The goal of the composite team is to get High School and Junior High boys and girls into mountain biking and racing. Even though cycling is an individual exercise...it is easily translated to a team sport. We support and encourage one another to grow and improve. My role is to help the newest riders gain confidence in a new endeavor and push themselves to experience the fun of mountain biking. Additionally, I am the team mechanic...so I regularly get greasy and dirty as the bikes are…
The goal of the composite team is to get High School and Junior High boys and girls into mountain biking and racing. Even though cycling is an individual exercise...it is easily translated to a team sport. We support and encourage one another to grow and improve. My role is to help the newest riders gain confidence in a new endeavor and push themselves to experience the fun of mountain biking. Additionally, I am the team mechanic...so I regularly get greasy and dirty as the bikes are (thankfully) ridden very hard.
Honors & Awards
-
Qualified for 2014 National Triathlon Championship - Olympic distance
USA Triathlon - usatriathlon.org
-
Qualified for 2014 Escape from Alcatraz Triathlon
IMG Worldwide
-
Qualified for 2012 Escape from Alcatraz Triathlon
IMG Worldwide
Languages
-
Spanish
Elementary proficiency
-
Ruby on Rails
Professional working proficiency
-
Python
Professional working proficiency
-
Go
Limited working proficiency
-
Rust
Elementary proficiency
-
Python
Professional working proficiency
-
Node.js and React.js.
Limited working proficiency
Recommendations received
10 people have recommended Benjamin
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content