Tadiwa Mbuwayesango

I started out studying Finance & Econ, but building ventures like @LEARN__X made me realize the future isn’t built in silos. Now I’m pursuing Computer Engineering at Iowa State, while minoring in business + econ, stacking tech depth on top of markets & strategy…

I saw a clip recently — “vibecoding during big tech interviews.” Fun in 2024 when Roy Lee went viral, sure. But honestly? You’re better off going deep on AI research. The roles are shifting fast: Software Engineer → AI Engineer → AI Research/Data Science. Everyone’s an “AI engineer” now. Meanwhile 24–26 y/o interns are landing spots at top labs just by binge-watching LLM training talks and building something interesting. Back in 2023–24 there weren’t even internship slots at these places. Moral of the story: Work your way toward becoming the next Karpathy. Sounds hard, but the competition is way thinner than you think.

Technically don’t need a phd in Math for sure. Most of the math you would have to understand is linear algebra and probability and a few statistics concepts but most of them you can learn while on the route of getting deep into AI. Also most AI research has really moved from papers to implementation.

@ElvisKimara Genuine question: what does “going deep on AI research” look like practically for someone without a math PhD — are you talking reproducing papers, reading and implementing architectures, or something else entirely?

@daniel_mac8 Don’t you think this is a more Pro model use case, you get better CoT and results there for this.

Codex Pro tip: turn Codex into a research engineer. Take any new agent paper and: 1. Ask: “How would this work in Codex?” 2. Enter /goal mode 3. Set the goal to implement it in your local config There’s a wave of research coming out on harness engineering: skills, memory, eval loops, tool use, scaffolds, agent feedback. This workflow lets you pull those ideas into your actual Codex environment instead of just bookmarking them. SkillOpt is a perfect example. It treats agent skills as trainable external state: run a rollout, edit the skill, validate, keep what works. In the paper, SkillOpt improved GPT-5.5 inside a Codex-style harness by +24.8 points over the no-skill baseline. That’s the loop you can implement today. It's already possible: research paper → Codex goal → local agent upgrade

Pretty sure uni already model this, but issue is that international students already pay a lot so increasing the tuition will reduce enrollment to such an extent that the price increase won’t offset that decline. On another note, pre-AI, it was also of national interest to have more international students for talent acquisitions (so did it really benefit schools as well — research etc)

Universities would make exponential profits if they abolished out-of-state tuition for Americans and just increased the tuition for international students.

@lennox_saint @steipete Also curious

@steipete Curious what your reading workflow in codex is like 👀

I'm late to the party, but cmux is great. github.com/manaflow-ai/cm… current split: codex mac app: knowledege work, learning, reading cmux + codex cli: coding

@Robert027R1 Yes that’s true, once you hit a certain range of diversification, your earnings from adding more indices is outweighed with the costs.

Why doesn’t someone create the S&P 5000 which includes the 5,000 biggest companies in America. Wouldn’t that outperform the S&P 500?

@TechnolgiaTalks Local models 👀

AI Token Subsidies Ending #ai #entrepreneurship #startup #claude #openai

@TechnolgiaTalks Is OpenAI making hardware devices?

AI on Your Phone? Future of Local Models Explained! #ai #entrepreneurship #startup #claude #openai

@TechnolgiaTalks Cursor has some good data for SWE and once you succeed in SE then the model unlocks limitless opportunities.

Should Elon Musk by Cursor? AI Chip Costs & Data Center Investments #ai #startup #claude #openai

@TechnolgiaTalks Apple investors would probably be happy for acquisitions😂

Steve Jobs & AI Hype: The Apple Acquisition Wars #ai #entrepreneurship #startup #claude #openai

@TechnolgiaTalks I rarely use my Claude code these days.

@TechnolgiaTalks Update, I gave it a pressure test and I still love it! Codex app has also been cooking!

Codex 5.5 Update: The Ultimate AI Coding Tool! #ai #entrepreneurship #startup #claude #openai

Video projects should have proper history. Most editing workflows still depend on one project file, an undo stack that disappears after you close the app, and filenames like final_v7_REAL_final.drp. vedit adds a version-control layer for video editing: - snapshot a timeline - branch to try a different cut - compare two versions in plain English - merge changes back when the edit works Instead of a wall of code, the diff reads like: - Moved b_roll_03 before interview_take_2 - Trimmed drone_shot_04 by 1.80 seconds - Added crossfade between title_card and drone_shot_04 The first release supports workflows around DaVinci Resolve, Premiere, Final Cut, but mostly AI-video editors. It is early, but the goal is clear: video editing should be collaborative, traceable, and easy to undo across days, exports, and teammates. PyPI: pypi.org/project/pyvedi… GitHub: github.com/explicit09/ved…

just shipped vedit — version control for video edits. snapshot timeline, branch, diff changes, merge. like git, but for cuts. the diffs read like English: Moved "b_roll_03" before "interview_take_2" Trimmed "drone_shot_04" by 1.80s (in) Added crossfade between "title_card" and "drone_shot_04" (12 frames) works on anything that exports OTIO — Resolve, Premiere, FCP, or AI tools that generate timelines. there's a vedit watch mode that auto-commits every time you export from Resolve. Rust core, Python bindings. very early (v0.0.1). would love eyes on it. PyPI: pypi.org/project/pyvedi… GitHub: github.com/explicit09/ved…

@ElvisKimara Definitely talk to @bcherny @trq212… it shouldn’t be that way!

Paid for Claude 2 weeks late after not using it at all. Next billing date stays the same, so I lose 2 weeks of paid service. How is this not a scam? I’m already paying $100 for the max plan. And basically, I’ve been billed for tokens I never used! Advice. If you are late to pay. Just pay the $20 for that month!

@elonmusk @wholemars @mark_k @ElvisKimara @TechnolgiaTalks

@mark_k We have many great Grok models training simultaneously in Colossus 2. I am confident that they will bear fruit. And work on the Grok Built harness is progressing well.

Retards are already announcing the death of Grok. 🤦‍♂️

Is Elon distracted. $60 billion for cursor isn’t worth it! #ai #sf #techtrends #claude #grok #openai

@TechnolgiaTalks Are we ever getting Grok Code?

Why can’t Elon build his own harness for AI instead of buying Cursor? #ai #sf #techtrends #claude #grok #openai

@TechnolgiaTalks Two good CEOs perfectly set for two different timelines of the company!

Steve Jobs vs Tim Cook? Who was a better CEO at Apple. #apple #ai #sf #techtrends #claude #grok #openai