Joshyspeed
Region: US
Tuesday 21 July 2026 21:22:48 GMT
38420
2113
103
95
Music
Download
Comments
user8807646359583 :
Anthropic fumbled so bad with the FUD around Mythos and Fable refuses to let me do anything because they labeled me as a biologist.
2026-07-22 04:35:57
131
diksha :
i think it’s important to consider the other side - anthropic has shown time and again that they care about ai safety. Banning powerful models to slow down AI development until regulation can catch up is not a bad idea. Labeling it as “lobbying the government for their personal interest” is biased.
2026-07-22 06:18:33
18
Mavrick :
SpaceX's core business model is lighting money on fire, so they're immune to competition threat
2026-07-22 12:40:37
32
John J W :
"SpaceX is doing a great job" but also SpaceX is funding it's founder's crusade to end democracy and promote fascism. is that part of the "great job" or just a free side effect?
2026-07-22 11:37:36
16
user3235268067213 :
Allowing China to steal / distill our frontier models is not a good long term solution for American competitiveness
2026-07-22 01:20:17
10
Gabriel 🇩🇴 :
They waste time and money on trying to ban competition when they could better use that to compete even further. Figure out how to not be behind. Corporations are sick
2026-07-22 04:07:41
13
mr.mjau1 :
Why I'm paying for 20x max on Claude, a big part is the harness. I have a bunch of MCPs an heavy tool-use, lots of subagent workflows. When I can get that with open-source, I may be thinking of switching
2026-07-22 11:51:10
0
Gaymudgeon :
Excellent takes. Almost completely agree. I think you misread Anthropic’s goal. For them the point isn’t protecting their business. They truly believe nobody else can be trusted. Also it’s not really a “bubble”. The margin on inference is very real.
2026-07-22 02:35:16
6
Tumor :
Deepseek claimed the same if not more and it was all a complete lie. Just a diffused American model with Chinese censorship
2026-07-22 02:17:52
1
Snsejetfne :
You can run the harness on non-Anthropic inference via OpenRouter or locally. But without ~2TB of unified memory you struggle to run a Fable-class model like Kimi K3 — quantization gets aggressive even on top-end machines, and you need a sharding layer like EXO to come close to Anthropic performance on non-enterprise appliances. Which is why most of Anthropic’s business — underpinning domain-dominant wrappers and harnesses — is the moat. Swapping Anthropic out of these vertical-specific B2B companies isn’t just changing a JSON API string in an .md file. Defense of the moat: The switching cost isn’t the endpoint, it’s the accumulated behavioral calibration. Vertical wrappers tune prompts, evals, guardrails, and agentic scaffolding against one model’s failure modes for months. Swap the model and every eval regresses; you’re re-QAing your entire product. Add that frontier-class open weights are effectively undeployable outside hyperscale — the quantized local versions are different models with different failure surfaces — and the “commodity inference” thesis collapses. The moat is model-specific reliability under agentic load, compounded by every harness built on it.
2026-07-22 18:37:25
0
kyrianbbop :
moat = microsoft building the platform where businesses are able to deploy models most effectively for the organization
2026-07-22 05:22:09
0
lmwts :
good analysis, rare
2026-07-22 17:00:38
0
El Chonco :
guys look at my ORIGINAL llm model it's just as good if not better than fable
2026-07-22 19:03:58
1
mike_diamonde :
It'd be interesting to see how the frontier model providers are siphoning off institutional knowledge. enterprise TOS says they dont train model off the data. but there are many other ways for them to reuse your data outside the narrow TOS terms. i don't see any transparency in this space.
2026-07-22 13:15:02
0
Optimus Climb :
Because the intelligence is in the architecture not just the model
2026-07-22 15:36:49
1
HalimawMagJakol :
Chatgpt is still leading the competition hah
2026-07-22 13:27:45
2
helpfulturtle23 :
Not sure I agree with the idea that Anthropic or even OpenAI for that matter would deviate significantly from their usage in organizations. If you’ve been around enterprise software long enough you know these companies have multi million dollar multi year contracts. Making these models sticky regardless of your model switching argument which is an architecture notion not necessarily economics
2026-07-22 00:49:31
6
its_that_one_ai_guy :
This was a well thought out analysis. IMO.. Ai native service companies, that’s the moat. Perform the service at a level of fidelity and cost efficiency that current incumbents can’t possibly match. That’s what I’m betting on.
2026-07-22 05:30:43
1
MrTikTokHead :
The saving grace is that Anthropic and OpenAI are building agentic layers into their harnesses that are incredibly hard to beat. Pi.dev is solid, but the Codex and Claude Code CLIs are still so good they genuinely amaze me. Run Hermes, Codex, PI, and Claude Code through Supacode and you feel like a god. lol.
2026-07-22 05:03:16
2
Phoenix :
OpenAI and SpaceX will implode before Anthropic
2026-07-22 05:15:26
1
davecampbell65 :
Was looking for this account. You’ve been automating for years now. Truly an OG
2026-07-22 00:15:10
1
mohg :
im so happy to see the begining of anthropics downfall. They really thought they had the market cornerd 😂
2026-07-22 14:53:32
0
Sam :
The closed source bubble is about to pop I’m so excited
2026-07-22 15:41:23
0
Dug the Doggo :
Realistically they are all gonna get real close to Trump over the next two years and try to get him to forbid people from using the Chinese models for some silly reason they will make up
2026-07-22 18:16:53
0
Le shark :
God I hope this pops the bubble
2026-07-22 20:44:41
0
To see more videos from user @joshyspeed, please go to the Tikwm
homepage.
