What local models do you recommend installing for offline use, such as when on an airplane traveling? I'm a complete beginner to local models.
It really just depends on what your hardware is. If you have the hardware capacity to run a Gemma 31-billion-parameter model, you should do that, but it's always a trade-off between tokens per second. You can figure a lot of this out by heading over to Llama or Hugging Face and clicking get started — I have a free account over there. Scroll down to models and you can click on virtually whatever model you want. This one here is 700 billion parameters, so you're not going to be able to run that. First, filter by local, because some of these have cloud offerings too. Then you'll see something like Bonsai has 2B, 4B, 9B, 27B. What you could try doing is setting up the 27B; if that doesn't work, move back to the 9B; if that doesn't work, move back to the 4B. Obviously that's a vision model, so it's probably not exactly what you want. A bunch of these Neotron ones are pretty good. Qwen 3.8 I think is basically state-of-the-art right now, so you could go 27B and see if that runs. A lot of people host quantized models on Hugging Face, so you could get a quantized Qwen. You can also get like Qwen 3.5 4B quantized — an older model, but because it's quantized it runs on significantly smaller hardware. The optimization reduces the model weights from 8 gigs to 5.2 GB on disk, meaning you can literally run this thing on an 8 GB computer. So figure out your computer and adjust accordingly.
I understand the concern about big model companies dominating, but there are two main types of improvements happening right now: hardware and software algorithmic improvements. Hardware advances mean larger server clusters and compute resources, which let companies train bigger models faster. Software improvements—quantization, inference hacks, and other algorithmic tricks—make models run faster and smaller. Because of these algorithmic gains, especially from teams that lack massive capital, we can run efficient models locally. For example, the Chinese models like GLM 5.2 are highly optimized and can be run on a consumer laptop: a 1‑bit quantized version runs on a MacBook with 256 GB of RAM. In the next six months we’ll likely be able to run comparable models on phones. So the fear of being locked out is overstated. The simplest way to get started is to download Ollama, a platform that lets you run open‑source models locally. You can pull models such as Ornith, GLM 5.2, or Kimmy K 2.7‑code, and many community members provide quantized versions that run on modest hardware. Even if a model is too large, you can pay a few dollars for a hosted service to help set it up, and you’ll be fine.
You want to wrestle a little control back from the big monolithic AI companies, and I can help with that. There are two kinds of improvement happening right now: hardware (massive data centers letting labs train bigger, and per the bitter lesson, generally smarter, models) and software/algorithmic improvements (inference optimizations, quantization, and other hacks that let you run faster or smaller models at similar intelligence). The hardware path isn't very accessible if you're not capital-rich — big models just cost more to run and labs will charge accordingly. But the algorithmic path doesn't require massive hardware. China, rate-limited on compute compared to the US's roughly 50-100x compute advantage, has poured its effort into this path — models like GLM 5.2 and Kimi K2 have gotten almost an order of magnitude more efficient purely from software optimization. As a result, you can run a model like GLM 5.2 locally right now, 1-bit quantized, on a MacBook with 256GB of memory. It's not unreasonable to expect that within about 6 months we'll be running something similar (or a bit dumber) on our phones. So while the fear of big model companies taking over is understandable, it seems very likely everyone will soon be able to run models much smarter than what we have today on compute as modest as a phone — you won't get the frontier models doing galaxy-scale planning, but you'll have genuinely capable local assistants you don't depend on any one company for.
local modelsopen source aivendor lock-inquantization
First, I'd recognize that where I'm physically located has virtually no bearing on how much money I can make — there's some impact if I'm from a region associated with offshore workers pitching people used to different price expectations, but assuming it's just me going to a new area, it doesn't really matter where I work or live. Second, I'd spend $184 of the $300 on Maker School — not because I need the info, but because the program bundles something like $20,000 in signup discounts and credits to AI providers. After the $184, I'm down to $116 but now have roughly $3,000 in equivalent credits and discounts, which is basically a 15x return before I've even used the program's actual content. Those credits use API pricing rather than plan pricing, so I'd sign up for something like a Claude Code or Codex plan (~$20/month), and once I burn through the usage limits there, fall back on the banked API credits and discounts, giving me effectively unlimited use of coding tools for at least a month — and if I need more beyond that, I'd sign up for something like Kimi K2 or DeepSeek's plan, which are less intelligent per dollar than a frontier model but plenty for the low-intelligence tasks I actually need. That leaves about $90. I'd put roughly $40 into an Instantly plan and another $40 into mailboxes, leaving about $16, then use a lead-gen platform like Apify (discounted through Maker School) to pull a list of ~15,000 leads. I'd rinse those leads through the Instantly mailboxes every day for the next month — realistically budgeting time for mailbox warm-up — and just focus on outreach daily: hundreds of individual touchpoints a day, religiously checking and responding to replies first thing every morning, sending Loom videos to anyone showing interest, following up relentlessly, and adding responders on LinkedIn and their YouTube to learn everything about them so I squeeze maximum value out of every lead — making my 300 leads perform like someone else's 1,000. Then I'd front-load as many sales calls as possible, offer reasonably low rates to lock in a first client and bootstrap credibility, and over-deliver heavily to maximize the odds of turning it into a retainer. The core insight: if you're even halfway current on this space's developments, you already know more than 99.9% of everyone else, and that's an extremely powerful arbitrage window right now — so join a discount aggregator (doesn't have to be Maker School, though obviously I think it's the best one), and flip as much of your starting money as possible into bootstrapping outreach.
You're sending about four proposals per day on average. If you increased that to 10 per day (2.5× more), you'd see roughly 2.5× more results. Your main bottleneck is the view rate, currently around 12 %; improving your CV, profile, how it stands out, and your writing to avoid generic templates could push that to 20‑25 %. Make sure the first 150 characters show you read the post, give something free, and present yourself as a professional. With a 12 % view rate, 16 viewed proposals translate to about five replies, which is rough but not hopeless. In my early days I sent 20‑25 proposals a day; I would have completed your 30‑day volume in about five and a half days, outperforming you six‑fold on views alone. Even without improving my view rate, that would have yielded roughly 90 views, around 30 replies, and three or four calls—enough to eventually land a deal. You can make up for any shortfall in volume with strategic effort, something I've done since day one. I'd still recommend you do it more effectively, but the principle holds.