Founder @Desearch_ai | SN22 Bittensor (Decentralized Search)
Joined November 2013
- Tweets 1,100
- Following 740
- Followers 576
- Likes 600
27 Photos and videos
Sounds like a good problem: a fast cache.
It is not good if stale cache blocks a real user with “insufficient balance”.
Reliability is knowing when not to trust your fast path.
The slow truth still needs a way back into the product.
40
Giga | Cosmic retweeted
Jun 13
Access to intelligence should not depend on a handful of companies or governments.
This is why open, decentralized, permissionless AI matters.
This is why Bittensor matters.
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
70
374
1,273
85,833
Jun 13
small API lesson:
a search result is not useful if it cannot power the next step.
If TikTok search returns an ID, a builder expects to fetch comments, inspect pain, and keep researching.
Tiny response contracts decide whether workflows compound or break.
26
Lol, Governments will try, but won't be able to stop/suppress AI innovation and growth. This is why distributed decentralised AI is so crucial...and probably why Bittensor got a bid today...
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
9
17
133
4,612
Jun 13
If I had to research a market from zero, I would not start with a spreadsheet.
I would ask:
1. who is already buying
2. what pain shows up in public
3. what tools they mention
4. which sources prove it
The source trail is the product.
30
Jun 12
SN22 scoring has a hard product lesson:
if two miners are different in quality, rewards should not pretend they are equal.
Small scoring curves decide whether a network learns.
Bad incentives make average answers cheap. Good incentives make better evidence worth chasing.
1
37
Jun 12
good Bittensor signal here.
search is becoming evaluation infrastructure, not just a user-facing box.
if miner outputs can be backtested, the subnet can reward behavior over vibes.
same principle for decentralized search: make quality inspectable.
Jun 11
Our backtestable search endpoint is now live for Numinous miners.
The endpoint integrates Numinous news sources, RSS feeds, geopolitics live data, base rates, and broader agentic retrieval into a growing research corpus.
This gives miners a stronger foundation for building forecasting agents on the subnet.
It also makes miner outputs easier to backtest, validate, and integrate into Numinous meta-models.
In the coming days, this search endpoint will become the required way to retrieve data on the subnet, making miner agents truly backtestable.
Search:
eversight.numinouslabs.io/do…
Fetch:
eversight.numinouslabs.io/do…
1
5
165
Giga | Cosmic retweeted
Desearch is showing long-term conviction onchain.
763.7K SN22 alpha is perpetually locked under Bittensor's Conviction upgrade, about 24.1% of staked SN22 alpha.
501K by Desearch.
262.7K by external holders.
Conviction for decentralized search.
#Bittensor #SN22 #Desearch
7
25
635
the best Bittensor conversations keep coming back to the same question:
what is the product proof?
not just emissions, not just demos.
who uses the output?
what improves when miners compete?
what can builders plug into today?
1
42
this is the part i care about most:
miners now need real coverage, not one easy lane.
AI search, X search, and Web search push different behavior.
if SN22 gets incentives right, the product gets better every time miners compete.
Subnet 22 validator update is merged.
Rewards now split 80% AI search, 10% X, 10% Web.
Each type has a quality bar, and real volume matters. Serving only one type no longer earns full credit.
PR: github.com/Desearch-ai/subne…
#Bittensor #SN22 #Desearch
1
63
multi-agent work breaks trust when child agents vanish.
if a subagent finished but the final message is empty, don't collapse it into “no output” too early.
recover the transcript if it exists. if it doesn't, say exactly what happened.
22
scheduled agents are only useful if they receive the exact task you approved.
one duplicated field can turn the real prompt into `[object Object]`.
boring lesson: for cron agents, the canonical payload matters more than the wrapper around it.
28
public eval data changes the conversation.
instead of “trust us, search is better”, anyone can inspect the questions, sources, scores, and verdicts.
for a subnet like SN22, that matters because incentives need to be debugged in public.
32
this is what i want more builders to understand about SN22.
not only “decentralized search” as an idea.
miners fetch live data, validators score quality, and builders get an API they can actually use.
that bridge from subnet to product is the hard part.
Latest article is live on IntoTAO: @desearch_ai
intotao.app/subnets/desearch
1
56
Skill install should feel easy, but the boundary should be strict.
resolve the exact commit
copy only the intended skill path
run install policy checks
test the flow end to end
Agent tooling needs one-click UX with boring supply-chain guardrails.
32
Good error messages are product work.
“provider unavailable” tells the user nothing.
“install this plugin or run doctor” turns the same failure into a next step.
The best fix is sometimes not more automation.
It is telling the user exactly how to recover.
33
this is the part of decentralized search that matters most to me:
not just “we have miners”.
the loop is:
miners compete
validators measure quality
rewards move toward better answers
if that loop keeps improving, search becomes a living product, not a static index.
The key idea behind decentralized search:
Miners compete to return better answers and better sources.
Validators measure quality.
Rewards move toward better behavior.
The product improves in public.
That is the loop we are building on Desearch.
#Bittensor #SN22
1
40
Question for founders building workflow tools:
when does a status update become motivating instead of noisy?
1. customer events
2. milestone trophies
3. team feed
4. weekly summary
I think progress works when it changes behavior, not when it just adds motion.
34
Good rule for agent file handling:
optional staging should not destroy the user’s input.
If a safe managed PDF can still be read, pass it through instead of failing the whole send.
Fallbacks are not just reliability.
They are part of user trust.
51
this is the Bittensor flywheel I care about:
one subnet makes compute easier.
another makes search better.
users should not need to see the plumbing, but builders should notice when subnet outputs start becoming real product dependencies.
Jun 3
Chutes Search is now powered by @desearch_ai (SN22), the decentralized search layer for agents and humans.
In their first public benchmark, it beat the centralized providers on overall score and ranked best of all of them at finding relevant sources. So we swapped it in to power our search.
Two subnets, one better search experience.
This is the Bittensor flywheel shipping.
SN22 x SN64. Want to try it?
search.chutes.ai/
4
133